Semantic Segmentationのいろんなモデルが扱えるgithubがあったので、それで、性能出ししてみる。DeepLabV3+とか。
目的
Semantic Segmentationのいろんなモデルが扱えるgithubがあったので、それで、性能出ししてみる。
該当のgithubは、以下。
https://github.com/luyanger1799/Amazing-Semantic-Segmentation
結果の前に。
下記の結果、その1、その2は、間違えています!!!
すみません。
Data Augmentation Rate --> 0.0
だと、オーグメンテーション、全く、作用していません。
ミス!!!!!!
(誰か、教えてよーーーー!)
立て直し中です、しばし。
結果
【無効】その1(オーグメンテーションの効果を探る)<--間違えています。上記のとおり。
まずは、この記事でのリファレンス。
設定は、以下。
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet50
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 8
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.0
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 5.0
Zoom --> [0.1]
Channel Shift --> 0.0
最初の10回。val_mean_io_u: 0.2916が最高。
Epoch 1/300
52/52 [==============================] - 61s 1s/step - loss: 89854.7969 - mean_io_u: 0.0889 - val_loss: 4838029.0000 - val_mean_io_u: 0.0375
Epoch 00002: LearningRateScheduler reducing learning rate to 0.00029999177814635854.
Epoch 2/300
52/52 [==============================] - 54s 1s/step - loss: 56099.2969 - mean_io_u: 0.1373 - val_loss: 626711.7500 - val_mean_io_u: 0.0610
Epoch 00003: LearningRateScheduler reducing learning rate to 0.00029996711348705303.
Epoch 3/300
52/52 [==============================] - 66s 1s/step - loss: 48637.7344 - mean_io_u: 0.1652 - val_loss: 584699.3125 - val_mean_io_u: 0.0639
Epoch 00004: LearningRateScheduler reducing learning rate to 0.0002999260087268414.
Epoch 4/300
52/52 [==============================] - 91s 2s/step - loss: 43062.8906 - mean_io_u: 0.1899 - val_loss: 82439.8672 - val_mean_io_u: 0.1594
Epoch 00005: LearningRateScheduler reducing learning rate to 0.0002998684683733238.
Epoch 5/300
52/52 [==============================] - 119s 2s/step - loss: 40139.8750 - mean_io_u: 0.2087 - val_loss: 48446.8047 - val_mean_io_u: 0.2208
Epoch 00006: LearningRateScheduler reducing learning rate to 0.0002997944987364483.
Epoch 6/300
52/52 [==============================] - 157s 3s/step - loss: 36975.4180 - mean_io_u: 0.2235 - val_loss: 48954.8555 - val_mean_io_u: 0.2482
Epoch 00007: LearningRateScheduler reducing learning rate to 0.00029970410792781927.
Epoch 7/300
52/52 [==============================] - 191s 4s/step - loss: 34668.2188 - mean_io_u: 0.2472 - val_loss: 42038.1367 - val_mean_io_u: 0.2570
Epoch 00008: LearningRateScheduler reducing learning rate to 0.00029959730585980745.
Epoch 8/300
52/52 [==============================] - 218s 4s/step - loss: 32943.8516 - mean_io_u: 0.2600 - val_loss: 37517.3281 - val_mean_io_u: 0.2845
Epoch 00009: LearningRateScheduler reducing learning rate to 0.00029947410424446315.
Epoch 9/300
52/52 [==============================] - 249s 5s/step - loss: 30667.6152 - mean_io_u: 0.2835 - val_loss: 36129.5820 - val_mean_io_u: 0.2520
Epoch 00010: LearningRateScheduler reducing learning rate to 0.00029933451659223183.
Epoch 10/300
52/52 [==============================] - 262s 5s/step - loss: 29595.3750 - mean_io_u: 0.2999 - val_loss: 36362.9062 - val_mean_io_u: 0.2916
末尾の10回。val_mean_io_u: 0.4908が最高。
Epoch 00291: LearningRateScheduler reducing learning rate to 9.214417895274202e-07.
Epoch 291/300
52/52 [==============================] - 361s 7s/step - loss: 6189.0298 - mean_io_u: 0.7719 - val_loss: 48670.3945 - val_mean_io_u: 0.4908
Epoch 00292: LearningRateScheduler reducing learning rate to 7.654834077681545e-07.
Epoch 292/300
52/52 [==============================] - 361s 7s/step - loss: 6234.5962 - mean_io_u: 0.7714 - val_loss: 38513.7070 - val_mean_io_u: 0.4161
Epoch 00293: LearningRateScheduler reducing learning rate to 6.258957555368196e-07.
Epoch 293/300
52/52 [==============================] - 361s 7s/step - loss: 6169.3179 - mean_io_u: 0.7722 - val_loss: 50993.7188 - val_mean_io_u: 0.3860
Epoch 00294: LearningRateScheduler reducing learning rate to 5.026941401925086e-07.
Epoch 294/300
52/52 [==============================] - 361s 7s/step - loss: 6170.3516 - mean_io_u: 0.7707 - val_loss: 53318.4062 - val_mean_io_u: 0.4438
Epoch 00295: LearningRateScheduler reducing learning rate to 3.958920721806797e-07.
Epoch 295/300
52/52 [==============================] - 361s 7s/step - loss: 6187.1567 - mean_io_u: 0.7709 - val_loss: 52796.4141 - val_mean_io_u: 0.4067
Epoch 00296: LearningRateScheduler reducing learning rate to 3.055012635516537e-07.
Epoch 296/300
52/52 [==============================] - 361s 7s/step - loss: 6196.3765 - mean_io_u: 0.7724 - val_loss: 46177.6641 - val_mean_io_u: 0.4249
Epoch 00297: LearningRateScheduler reducing learning rate to 2.3153162667618543e-07.
Epoch 297/300
52/52 [==============================] - 361s 7s/step - loss: 6186.9629 - mean_io_u: 0.7720 - val_loss: 41963.2383 - val_mean_io_u: 0.4324
Epoch 00298: LearningRateScheduler reducing learning rate to 1.7399127315854657e-07.
Epoch 298/300
52/52 [==============================] - 361s 7s/step - loss: 6207.0586 - mean_io_u: 0.7711 - val_loss: 53273.1016 - val_mean_io_u: 0.4153
Epoch 00299: LearningRateScheduler reducing learning rate to 1.3288651294691732e-07.
Epoch 299/300
52/52 [==============================] - 361s 7s/step - loss: 6169.7905 - mean_io_u: 0.7702 - val_loss: 45624.3047 - val_mean_io_u: 0.4071
Epoch 00300: LearningRateScheduler reducing learning rate to 1.0822185364145413e-07.
Epoch 300/300
52/52 [==============================] - 361s 7s/step - loss: 6178.2095 - mean_io_u: 0.7718 - val_loss: 51067.3398 - val_mean_io_u: 0.3722
【無効】その2(オーグメンテーションを増やす)<--間違えています。上記のとおり。
設定は、以下。
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet50
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 8
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.0
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 10.0★
Zoom --> [0.2]★
Channel Shift --> 0.1★
最初の10回。val_mean_io_u: 0.3243が最高。好スタート、たまたま?。
Epoch 1/300
52/52 [==============================] - 63s 1s/step - loss: 89454.6719 - mean_io_u: 0.0912 - val_loss: 3623744.7500 - val_mean_io_u: 0.0438
Epoch 00002: LearningRateScheduler reducing learning rate to 0.00029999177814635854.
Epoch 2/300
52/52 [==============================] - 92s 2s/step - loss: 56037.2383 - mean_io_u: 0.1385 - val_loss: 448488.5000 - val_mean_io_u: 0.0565
Epoch 00003: LearningRateScheduler reducing learning rate to 0.00029996711348705303.
Epoch 3/300
52/52 [==============================] - 343s 7s/step - loss: 47359.6797 - mean_io_u: 0.1658 - val_loss: 113093.2969 - val_mean_io_u: 0.1230
Epoch 00004: LearningRateScheduler reducing learning rate to 0.0002999260087268414.
Epoch 4/300
52/52 [==============================] - 363s 7s/step - loss: 43799.0898 - mean_io_u: 0.1870 - val_loss: 64006.3242 - val_mean_io_u: 0.1854
Epoch 00005: LearningRateScheduler reducing learning rate to 0.0002998684683733238.
Epoch 5/300
52/52 [==============================] - 363s 7s/step - loss: 39738.7617 - mean_io_u: 0.2064 - val_loss: 53173.1133 - val_mean_io_u: 0.2211
Epoch 00006: LearningRateScheduler reducing learning rate to 0.0002997944987364483.
Epoch 6/300
52/52 [==============================] - 363s 7s/step - loss: 37022.7500 - mean_io_u: 0.2232 - val_loss: 42792.3008 - val_mean_io_u: 0.2476
Epoch 00007: LearningRateScheduler reducing learning rate to 0.00029970410792781927.
Epoch 7/300
52/52 [==============================] - 362s 7s/step - loss: 36197.5508 - mean_io_u: 0.2375 - val_loss: 48736.1406 - val_mean_io_u: 0.2034
Epoch 00008: LearningRateScheduler reducing learning rate to 0.00029959730585980745.
Epoch 8/300
52/52 [==============================] - 364s 7s/step - loss: 33251.0586 - mean_io_u: 0.2575 - val_loss: 45869.1484 - val_mean_io_u: 0.2790
Epoch 00009: LearningRateScheduler reducing learning rate to 0.00029947410424446315.
Epoch 9/300
52/52 [==============================] - 363s 7s/step - loss: 31285.2715 - mean_io_u: 0.2837 - val_loss: 41630.4805 - val_mean_io_u: 0.3243
Epoch 00010: LearningRateScheduler reducing learning rate to 0.00029933451659223183.
Epoch 10/300
52/52 [==============================] - 361s 7s/step - loss: 29771.8164 - mean_io_u: 0.3023 - val_loss: 34137.9453 - val_mean_io_u: 0.2640
その3(オーグメンテーションを入れる)
設定は、以下。
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet50
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 8
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5★
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 10.0★
Zoom --> 0.2★
Channel Shift --> 0.1
様子は、以下。
Epoch 79/300
52/52 [==] - 360s 7s/step - loss: 18545.3730 - mean_io_u: 0.4889 - val_loss: 47864.3594 - val_mean_io_u: 0.3305
Epoch 00080: LearningRateScheduler reducing learning rate to 0.00025154755165391495.
Epoch 80/300
52/52 [==] - 360s 7s/step - loss: 18201.3906 - mean_io_u: 0.4930 - val_loss: 31374.0371 - val_mean_io_u: 0.4434
Epoch 00081: LearningRateScheduler reducing learning rate to 0.00025038613442351075.
Epoch 81/300
52/52 [==] - 360s 7s/step - loss: 19358.6328 - mean_io_u: 0.4808 - val_loss: 25203.5059 - val_mean_io_u: 0.4636
Epoch 00082: LearningRateScheduler reducing learning rate to 0.0002492137142052816.
Epoch 82/300
52/52 [==] - 360s 7s/step - loss: 18497.4629 - mean_io_u: 0.4865 - val_loss: 37067.1367 - val_mean_io_u: 0.3622
Epoch 00083: LearningRateScheduler reducing learning rate to 0.00024803041956831634.
Epoch 83/300
52/52 [==] - 360s 7s/step - loss: 20073.7539 - mean_io_u: 0.4628 - val_loss: 40759.3125 - val_mean_io_u: 0.4400
Epoch 00084: LearningRateScheduler reducing learning rate to 0.0002468363802742064.
Epoch 84/300
52/52 [==] - 360s 7s/step - loss: 17822.4258 - mean_io_u: 0.4938 - val_loss: 34194.4219 - val_mean_io_u: 0.3999
Epoch 00085: LearningRateScheduler reducing learning rate to 0.000245631727262816.
Epoch 85/300
52/52 [==] - 363s 7s/step - loss: 18174.8242 - mean_io_u: 0.4932 - val_loss: 33820.6367 - val_mean_io_u: 0.3827
Epoch 00086: LearningRateScheduler reducing learning rate to 0.0002444165926379231.
Epoch 86/300
52/52 [==] - 360s 7s/step - loss: 19058.8965 - mean_io_u: 0.4800 - val_loss: 35308.7734 - val_mean_io_u: 0.4374
悪くはない。
その4(オーグメンテーションを入れる。少し値をかえる。)
設定は、以下。
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet50
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 8
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 5.0
Zoom --> 0.1
Channel Shift --> 0.1
様子は、以下。
Epoch 00117: LearningRateScheduler reducing learning rate to 0.0002023183784959062.
Epoch 117/300
52/52 [==] - 363s 7s/step - loss: 16876.9238 - mean_io_u: 0.5123 - val_loss: 32107.6934 - val_mean_io_u: 0.4336
Epoch 00118: LearningRateScheduler reducing learning rate to 0.00020084375114078147.
Epoch 118/300
52/52 [==] - 362s 7s/step - loss: 15673.7490 - mean_io_u: 0.5248 - val_loss: 36437.9805 - val_mean_io_u: 0.4470
Epoch 00119: LearningRateScheduler reducing learning rate to 0.00019936355367845054.
Epoch 119/300
52/52 [==] - 362s 7s/step - loss: 15959.3994 - mean_io_u: 0.5219 - val_loss: 40523.1992 - val_mean_io_u: 0.4580
Epoch 00120: LearningRateScheduler reducing learning rate to 0.0001978779484292456.
Epoch 120/300
52/52 [==] - 362s 7s/step - loss: 15329.0020 - mean_io_u: 0.5312 - val_loss: 43972.2344 - val_mean_io_u: 0.3720
Epoch 00121: LearningRateScheduler reducing learning rate to 0.00019638709830652336.
Epoch 121/300
52/52 [==] - 362s 7s/step - loss: 15349.1074 - mean_io_u: 0.5255 - val_loss: 36461.8984 - val_mean_io_u: 0.4222
Epoch 00122: LearningRateScheduler reducing learning rate to 0.00019489116679880018.
Epoch 122/300
52/52 [==] - 362s 7s/step - loss: 15965.5615 - mean_io_u: 0.5266 - val_loss: 27753.3066 - val_mean_io_u: 0.4364
Epoch 00123: LearningRateScheduler reducing learning rate to 0.0001933903179518235.
Epoch 123/300
52/52 [==] - 363s 7s/step - loss: 16275.4512 - mean_io_u: 0.5250 - val_loss: 23859.1914 - val_mean_io_u: 0.4963
Epoch 00124: LearningRateScheduler reducing learning rate to 0.00019188471635058244.
Epoch 124/300
52/52 [==] - 363s 7s/step - loss: 15704.1211 - mean_io_u: 0.5308 - val_loss: 34267.9141 - val_mean_io_u: 0.3939
Epoch 00125: LearningRateScheduler reducing learning rate to 0.00019037452710125906.
Epoch 125/300
52/52 [==] - 363s 7s/step - loss: 15373.4512 - mean_io_u: 0.5292 - val_loss: 37838.2383 - val_mean_io_u: 0.4265
Epoch 00126: LearningRateScheduler reducing learning rate to 0.000188859915813123.
Epoch 126/300
52/52 [==] - 362s 7s/step - loss: 16315.5293 - mean_io_u: 0.5243 - val_loss: 32161.3086 - val_mean_io_u: 0.4616
Epoch 00127: LearningRateScheduler reducing learning rate to 0.00018734104858036995.
Epoch 127/300
52/52 [==] - 363s 7s/step - loss: 16153.8672 - mean_io_u: 0.5211 - val_loss: 31863.3027 - val_mean_io_u: 0.4547
Epoch 00128: LearningRateScheduler reducing learning rate to 0.0001858180919639082.
Epoch 128/300
52/52 [==] - 362s 7s/step - loss: 15989.2305 - mean_io_u: 0.5274 - val_loss: 31548.3965 - val_mean_io_u: 0.4523
Epoch 00129: LearningRateScheduler reducing learning rate to 0.00018429121297309283.
Epoch 129/300
52/52 [==] - 362s 7s/step - loss: 16574.8555 - mean_io_u: 0.5251 - val_loss: 36970.0547 - val_mean_io_u: 0.4592
Epoch 00130: LearningRateScheduler reducing learning rate to 0.00018276057904741157.
Epoch 130/300
52/52 [==] - 363s 7s/step - loss: 15560.6650 - mean_io_u: 0.5310 - val_loss: 28971.3867 - val_mean_io_u: 0.4283
Epoch 00131: LearningRateScheduler reducing learning rate to 0.000181226358038123.
Epoch 131/300
52/52 [==] - 362s 7s/step - loss: 15101.3955 - mean_io_u: 0.5385 - val_loss: 42130.1641 - val_mean_io_u: 0.4547
Epoch 00132: LearningRateScheduler reducing learning rate to 0.00017968871818984994.
Epoch 132/300
52/52 [==] - 363s 7s/step - loss: 14920.5439 - mean_io_u: 0.5449 - val_loss: 35276.6406 - val_mean_io_u: 0.4440
Epoch 00133: LearningRateScheduler reducing learning rate to 0.0001781478281221294.
Epoch 133/300
52/52 [==] - 362s 7s/step - loss: 15017.3975 - mean_io_u: 0.5361 - val_loss: 31138.0371 - val_mean_io_u: 0.4209
悪くはない。ただ、結果3との差は、そんなたいしたことない。
その5(ResNetを少し値をかえる。)
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet101
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 4
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 5.0
Zoom --> 0.1
Channel Shift --> 0.1
Epoch 51/300
105/105 [==] - 439s 4s/step - loss: 20754.2754 - mean_io_u: 0.4675 - val_loss: 35808.4609 - val_mean_io_u: 0.3351
Epoch 00052: LearningRateScheduler reducing learning rate to 0.0002791182669492413.
Epoch 52/300
105/105 [==] - 439s 4s/step - loss: 21148.6484 - mean_io_u: 0.4660 - val_loss: 24649.5918 - val_mean_io_u: 0.4525
Epoch 00053: LearningRateScheduler reducing learning rate to 0.00027831187081106795.
Epoch 53/300
105/105 [==] - 439s 4s/step - loss: 21197.6836 - mean_io_u: 0.4637 - val_loss: 38648.6953 - val_mean_io_u: 0.4103
Epoch 00054: LearningRateScheduler reducing learning rate to 0.0002774914093133802.
Epoch 54/300
105/105 [==] - 439s 4s/step - loss: 21719.2695 - mean_io_u: 0.4615 - val_loss: 36425.7266 - val_mean_io_u: 0.3813
Epoch 00055: LearningRateScheduler reducing learning rate to 0.00027665697242902713.
Epoch 55/300
105/105 [==] - 439s 4s/step - loss: 20564.6484 - mean_io_u: 0.4682 - val_loss: 27104.0098 - val_mean_io_u: 0.3563
Epoch 00056: LearningRateScheduler reducing learning rate to 0.0002758086516634163.
Epoch 56/300
105/105 [==] - 439s 4s/step - loss: 21082.5332 - mean_io_u: 0.4665 - val_loss: 36071.4336 - val_mean_io_u: 0.3828
Epoch 00057: LearningRateScheduler reducing learning rate to 0.0002749465400444794.
Epoch 57/300
105/105 [==] - 439s 4s/step - loss: 20906.1387 - mean_io_u: 0.4670 - val_loss: 29424.6289 - val_mean_io_u: 0.3872
Epoch 00058: LearningRateScheduler reducing learning rate to 0.0002740707321124705.
Epoch 58/300
105/105 [==] - 439s 4s/step - loss: 20159.6309 - mean_io_u: 0.4746 - val_loss: 32868.6758 - val_mean_io_u: 0.4212
Epoch 00059: LearningRateScheduler reducing learning rate to 0.0002731813239095989.
Epoch 59/300
105/105 [==] - 439s 4s/step - loss: 30735.2520 - mean_io_u: 0.3735 - val_loss: 100592.9219 - val_mean_io_u: 0.2188
Epoch 00060: LearningRateScheduler reducing learning rate to 0.00027227841296949664.
Epoch 60/300
105/105 [==] - 439s 4s/step - loss: 23701.4707 - mean_io_u: 0.4360 - val_loss: 45182.4414 - val_mean_io_u: 0.3620
Epoch 00061: LearningRateScheduler reducing learning rate to 0.00027136209830652333.
Epoch 61/300
105/105 [==] - 439s 4s/step - loss: 22356.2168 - mean_io_u: 0.4572 - val_loss: 35422.8008 - val_mean_io_u: 0.4299
Epoch 00062: LearningRateScheduler reducing learning rate to 0.0002704324804049076.
Epoch 62/300
105/105 [==] - 439s 4s/step - loss: 20632.4336 - mean_io_u: 0.4744 - val_loss: 21433.0664 - val_mean_io_u: 0.4589
Epoch 00063: LearningRateScheduler reducing learning rate to 0.0002694896612077282.
Epoch 63/300
105/105 [==] - 439s 4s/step - loss: 20519.9043 - mean_io_u: 0.4735 - val_loss: 31553.0039 - val_mean_io_u: 0.4271
Epoch 00064: LearningRateScheduler reducing learning rate to 0.00026853374410573476.
Epoch 64/300
105/105 [==] - 439s 4s/step - loss: 20935.2773 - mean_io_u: 0.4713 - val_loss: 28307.8750 - val_mean_io_u: 0.4411
Epoch 00065: LearningRateScheduler reducing learning rate to 0.00026756483392600966.
Epoch 65/300
105/105 [==] - 439s 4s/step - loss: 19320.7598 - mean_io_u: 0.4790 - val_loss: 32983.2422 - val_mean_io_u: 0.3830
Epoch 00066: LearningRateScheduler reducing learning rate to 0.0002665830369204728.
Epoch 66/300
105/105 [==] - 439s 4s/step - loss: 19557.0469 - mean_io_u: 0.4836 - val_loss: 28492.5684 - val_mean_io_u: 0.4533
Epoch 00067: LearningRateScheduler reducing learning rate to 0.00026558846075422956.
Epoch 67/300
105/105 [==] - 439s 4s/step - loss: 18712.0879 - mean_io_u: 0.4891 - val_loss: 28251.6973 - val_mean_io_u: 0.4499
Epoch 00068: LearningRateScheduler reducing learning rate to 0.0002645812144937646.
Epoch 68/300
105/105 [==] - 442s 4s/step - loss: 19172.1816 - mean_io_u: 0.4865 - val_loss: 33057.2656 - val_mean_io_u: 0.4645
Epoch 00069: LearningRateScheduler reducing learning rate to 0.00026356140859498087.
Epoch 69/300
105/105 [==] - 439s 4s/step - loss: 18947.2773 - mean_io_u: 0.4874 - val_loss: 27134.8066 - val_mean_io_u: 0.4332
Epoch 00070: LearningRateScheduler reducing learning rate to 0.0002625291548910874.
Epoch 70/300
105/105 [==] - 439s 4s/step - loss: 19917.4180 - mean_io_u: 0.4809 - val_loss: 30916.6211 - val_mean_io_u: 0.4319
Epoch 00071: LearningRateScheduler reducing learning rate to 0.00026148456658033525.
Epoch 71/300
105/105 [==] - 441s 4s/step - loss: 19324.2188 - mean_io_u: 0.4826 - val_loss: 23521.9668 - val_mean_io_u: 0.4877
Epoch 00072: LearningRateScheduler reducing learning rate to 0.00026042775821360413.
Epoch 72/300
105/105 [==] - 439s 4s/step - loss: 18859.3301 - mean_io_u: 0.4872 - val_loss: 28276.8438 - val_mean_io_u: 0.4266
Epoch 00073: LearningRateScheduler reducing learning rate to 0.00025935884568184064.
Epoch 73/300
105/105 [==] - 439s 4s/step - loss: 19002.8145 - mean_io_u: 0.4871 - val_loss: 32405.9004 - val_mean_io_u: 0.4064
Epoch 00074: LearningRateScheduler reducing learning rate to 0.0002582779462033494.
Epoch 74/300
105/105 [==] - 439s 4s/step - loss: 20063.4043 - mean_io_u: 0.4799 - val_loss: 61415.6641 - val_mean_io_u: 0.2587
Epoch 00075: LearningRateScheduler reducing learning rate to 0.0002571851783109388.
Epoch 75/300
105/105 [==] - 439s 4s/step - loss: 20974.9668 - mean_io_u: 0.4674 - val_loss: 37915.5391 - val_mean_io_u: 0.3546
Epoch 00076: LearningRateScheduler reducing learning rate to 0.00025608066183892275.
Epoch 76/300
105/105 [==] - 439s 4s/step - loss: 19704.2754 - mean_io_u: 0.4837 - val_loss: 28454.9336 - val_mean_io_u: 0.4140
Epoch 00077: LearningRateScheduler reducing learning rate to 0.00025496451790997917.
Epoch 77/300
105/105 [==] - 442s 4s/step - loss: 19353.9805 - mean_io_u: 0.4883 - val_loss: 29147.9941 - val_mean_io_u: 0.4403
Epoch 00078: LearningRateScheduler reducing learning rate to 0.00025383686892186747.
Epoch 78/300
105/105 [==] - 439s 4s/step - loss: 18555.5781 - mean_io_u: 0.4943 - val_loss: 33496.7188 - val_mean_io_u: 0.4315
Epoch 00079: LearningRateScheduler reducing learning rate to 0.00025269783853400685.
Epoch 79/300
105/105 [==] - 439s 4s/step - loss: 18904.0332 - mean_io_u: 0.4918 - val_loss: 34029.1016 - val_mean_io_u: 0.3891
Epoch 00080: LearningRateScheduler reducing learning rate to 0.00025154755165391495.
Epoch 80/300
105/105 [==] - 439s 4s/step - loss: 18675.8184 - mean_io_u: 0.4976 - val_loss: 31313.4941 - val_mean_io_u: 0.3922
Epoch 00081: LearningRateScheduler reducing learning rate to 0.00025038613442351075.
Epoch 81/300
105/105 [==] - 439s 4s/step - loss: 18792.6855 - mean_io_u: 0.4952 - val_loss: 30992.4316 - val_mean_io_u: 0.4525
Epoch 00082: LearningRateScheduler reducing learning rate to 0.0002492137142052816.
Epoch 82/300
105/105 [==] - 439s 4s/step - loss: 18501.2676 - mean_io_u: 0.5000 - val_loss: 25794.7461 - val_mean_io_u: 0.4364
Epoch 00083: LearningRateScheduler reducing learning rate to 0.00024803041956831634.
Epoch 83/300
105/105 [==] - 439s 4s/step - loss: 19045.1777 - mean_io_u: 0.4892 - val_loss: 32127.4961 - val_mean_io_u: 0.4550
Epoch 00084: LearningRateScheduler reducing learning rate to 0.0002468363802742064.
Epoch 84/300
105/105 [==] - 439s 4s/step - loss: 18327.3652 - mean_io_u: 0.5005 - val_loss: 27739.3125 - val_mean_io_u: 0.4778
Epoch 00085: LearningRateScheduler reducing learning rate to 0.000245631727262816.
Epoch 85/300
105/105 [==] - 439s 4s/step - loss: 17754.4766 - mean_io_u: 0.5002 - val_loss: 28532.9180 - val_mean_io_u: 0.4875
Epoch 00086: LearningRateScheduler reducing learning rate to 0.0002444165926379231.
Epoch 86/300
105/105 [==] - 439s 4s/step - loss: 17839.9180 - mean_io_u: 0.5031 - val_loss: 30797.6836 - val_mean_io_u: 0.4736
Epoch 00087: LearningRateScheduler reducing learning rate to 0.00024319110965273264.
Epoch 87/300
105/105 [==] - 439s 4s/step - loss: 18355.6367 - mean_io_u: 0.4975 - val_loss: 40958.4688 - val_mean_io_u: 0.4513
Epoch 00088: LearningRateScheduler reducing learning rate to 0.0002419554126952638.
Epoch 88/300
105/105 [==] - 439s 4s/step - loss: 18732.6699 - mean_io_u: 0.4837 - val_loss: 27917.7129 - val_mean_io_u: 0.4678
Epoch 00089: LearningRateScheduler reducing learning rate to 0.00024070963727361311.
Epoch 89/300
105/105 [==] - 439s 4s/step - loss: 16974.5566 - mean_io_u: 0.5161 - val_loss: 27659.2441 - val_mean_io_u: 0.4578
Epoch 00090: LearningRateScheduler reducing learning rate to 0.00023945392000109406.
Epoch 90/300
105/105 [==] - 439s 4s/step - loss: 17880.8027 - mean_io_u: 0.5086 - val_loss: 33514.3477 - val_mean_io_u: 0.4030
Epoch 00091: LearningRateScheduler reducing learning rate to 0.00023818839858125631.
Epoch 91/300
105/105 [==] - 439s 4s/step - loss: 17840.9590 - mean_io_u: 0.5088 - val_loss: 37918.2656 - val_mean_io_u: 0.3954
Epoch 00092: LearningRateScheduler reducing learning rate to 0.00023691321179278464.
Epoch 92/300
105/105 [==] - 439s 4s/step - loss: 18588.4316 - mean_io_u: 0.5025 - val_loss: 40244.8008 - val_mean_io_u: 0.4129
Epoch 00093: LearningRateScheduler reducing learning rate to 0.00023562849947428055.
Epoch 93/300
105/105 [==] - 439s 4s/step - loss: 19025.6387 - mean_io_u: 0.4988 - val_loss: 29079.4004 - val_mean_io_u: 0.4428
Epoch 00094: LearningRateScheduler reducing learning rate to 0.00023433440250892698.
Epoch 94/300
105/105 [==] - 441s 4s/step - loss: 17575.1172 - mean_io_u: 0.5111 - val_loss: 29741.3945 - val_mean_io_u: 0.4217
Epoch 00095: LearningRateScheduler reducing learning rate to 0.00023303106280903944.
Epoch 95/300
105/105 [==] - 439s 4s/step - loss: 17676.0742 - mean_io_u: 0.5118 - val_loss: 39939.6328 - val_mean_io_u: 0.4386
Epoch 00096: LearningRateScheduler reducing learning rate to 0.00023171862330050329.
Epoch 96/300
105/105 [==] - 441s 4s/step - loss: 17121.9590 - mean_io_u: 0.5116 - val_loss: 29737.2188 - val_mean_io_u: 0.5074
Epoch 00097: LearningRateScheduler reducing learning rate to 0.00023039722790710052.
Epoch 97/300
105/105 [==] - 439s 4s/step - loss: 16902.2910 - mean_io_u: 0.5219 - val_loss: 40803.8086 - val_mean_io_u: 0.4419
Epoch 00098: LearningRateScheduler reducing learning rate to 0.0002290670215347268.
Epoch 98/300
105/105 [==] - 439s 4s/step - loss: 18266.8066 - mean_io_u: 0.5088 - val_loss: 38198.0547 - val_mean_io_u: 0.4305
Epoch 00099: LearningRateScheduler reducing learning rate to 0.00022772815005550086.
Epoch 99/300
105/105 [==] - 442s 4s/step - loss: 17128.5957 - mean_io_u: 0.5155 - val_loss: 25639.7383 - val_mean_io_u: 0.4487
Epoch 00100: LearningRateScheduler reducing learning rate to 0.00022638076029176817.
Epoch 100/300
105/105 [==] - 439s 4s/step - loss: 18106.3809 - mean_io_u: 0.5102 - val_loss: 32126.6250 - val_mean_io_u: 0.5040
Epoch 00101: LearningRateScheduler reducing learning rate to 0.00022502499999999998.
Epoch 101/300
105/105 [==] - 439s 4s/step - loss: 16706.1816 - mean_io_u: 0.5222 - val_loss: 26992.8965 - val_mean_io_u: 0.4040
Epoch 00102: LearningRateScheduler reducing learning rate to 0.00022366101785459034.
Epoch 102/300
105/105 [==] - 439s 4s/step - loss: 16880.5645 - mean_io_u: 0.5211 - val_loss: 33327.0391 - val_mean_io_u: 0.4451
Epoch 00103: LearningRateScheduler reducing learning rate to 0.00022228896343155218.
Epoch 103/300
105/105 [==] - 439s 4s/step - loss: 17056.1016 - mean_io_u: 0.5218 - val_loss: 26874.1836 - val_mean_io_u: 0.4479
Epoch 00104: LearningRateScheduler reducing learning rate to 0.00022090898719211464.
Epoch 104/300
105/105 [==] - 439s 4s/step - loss: 17030.3848 - mean_io_u: 0.5206 - val_loss: 33408.6328 - val_mean_io_u: 0.4553
Epoch 00105: LearningRateScheduler reducing learning rate to 0.00021952124046622327.
Epoch 105/300
105/105 [==] - 439s 4s/step - loss: 17607.9121 - mean_io_u: 0.5163 - val_loss: 31780.9844 - val_mean_io_u: 0.4763
Epoch 00106: LearningRateScheduler reducing learning rate to 0.000218125875435945.
Epoch 106/300
105/105 [==] - 439s 4s/step - loss: 18413.9883 - mean_io_u: 0.5059 - val_loss: 29617.6406 - val_mean_io_u: 0.4220
Epoch 00107: LearningRateScheduler reducing learning rate to 0.00021672304511877985.
Epoch 107/300
105/105 [==] - 439s 4s/step - loss: 17377.0508 - mean_io_u: 0.5217 - val_loss: 32673.9883 - val_mean_io_u: 0.4035
Epoch 00108: LearningRateScheduler reducing learning rate to 0.0002153129033508805.
Epoch 108/300
105/105 [==] - 439s 4s/step - loss: 16938.3281 - mean_io_u: 0.5204 - val_loss: 30095.5469 - val_mean_io_u: 0.4556
Epoch 00109: LearningRateScheduler reducing learning rate to 0.00021389560477018263.
Epoch 109/300
105/105 [==] - 439s 4s/step - loss: 16809.3828 - mean_io_u: 0.5247 - val_loss: 42255.5508 - val_mean_io_u: 0.3945
Epoch 00110: LearningRateScheduler reducing learning rate to 0.00021247130479944715.
Epoch 110/300
105/105 [==] - 439s 4s/step - loss: 16895.0078 - mean_io_u: 0.5218 - val_loss: 34139.1914 - val_mean_io_u: 0.4332
Epoch 00111: LearningRateScheduler reducing learning rate to 0.00021104015962921622.
Epoch 111/300
105/105 [==] - 439s 4s/step - loss: 17304.8828 - mean_io_u: 0.5222 - val_loss: 30055.7598 - val_mean_io_u: 0.4552
Epoch 00112: LearningRateScheduler reducing learning rate to 0.0002096023262006853.
Epoch 112/300
105/105 [==] - 439s 4s/step - loss: 17113.0742 - mean_io_u: 0.5180 - val_loss: 50182.1328 - val_mean_io_u: 0.3320
Epoch 00113: LearningRateScheduler reducing learning rate to 0.00020815796218849283.
Epoch 113/300
105/105 [==] - 439s 4s/step - loss: 17079.6914 - mean_io_u: 0.5191 - val_loss: 28678.8125 - val_mean_io_u: 0.5060
Epoch 00114: LearningRateScheduler reducing learning rate to 0.00020670722598342914.
Epoch 114/300
105/105 [==] - 439s 4s/step - loss: 16533.6133 - mean_io_u: 0.5296 - val_loss: 36502.8125 - val_mean_io_u: 0.4180
Epoch 00115: LearningRateScheduler reducing learning rate to 0.00020525027667506746.
Epoch 115/300
105/105 [==] - 439s 4s/step - loss: 16671.6680 - mean_io_u: 0.5281 - val_loss: 31694.0723 - val_mean_io_u: 0.4586
Epoch 00116: LearningRateScheduler reducing learning rate to 0.00020378727403431773.
Epoch 116/300
105/105 [==] - 439s 4s/step - loss: 16059.1523 - mean_io_u: 0.5346 - val_loss: 36608.8867 - val_mean_io_u: 0.4365
Epoch 00117: LearningRateScheduler reducing learning rate to 0.0002023183784959062.
Epoch 117/300
105/105 [==] - 439s 4s/step - loss: 16285.6836 - mean_io_u: 0.5314 - val_loss: 34845.7578 - val_mean_io_u: 0.4614
Epoch 00118: LearningRateScheduler reducing learning rate to 0.00020084375114078147.
Epoch 118/300
105/105 [==] - 439s 4s/step - loss: 17148.4414 - mean_io_u: 0.5179 - val_loss: 31063.0938 - val_mean_io_u: 0.4196
Epoch 00119: LearningRateScheduler reducing learning rate to 0.00019936355367845054.
Epoch 119/300
105/105 [==] - 439s 4s/step - loss: 18019.8027 - mean_io_u: 0.5157 - val_loss: 30122.4219 - val_mean_io_u: 0.4677
Epoch 00120: LearningRateScheduler reducing learning rate to 0.0001978779484292456.
Epoch 120/300
105/105 [==] - 439s 4s/step - loss: 17251.8750 - mean_io_u: 0.5203 - val_loss: 32635.9121 - val_mean_io_u: 0.4971
Epoch 00121: LearningRateScheduler reducing learning rate to 0.00019638709830652336.
Epoch 121/300
105/105 [==] - 439s 4s/step - loss: 16250.2646 - mean_io_u: 0.5269 - val_loss: 36267.1016 - val_mean_io_u: 0.4771
Epoch 00122: LearningRateScheduler reducing learning rate to 0.00019489116679880018.
Epoch 122/300
105/105 [==] - 442s 4s/step - loss: 16181.2207 - mean_io_u: 0.5300 - val_loss: 22203.4297 - val_mean_io_u: 0.5335
Epoch 00123: LearningRateScheduler reducing learning rate to 0.0001933903179518235.
Epoch 123/300
105/105 [==] - 442s 4s/step - loss: 16958.2773 - mean_io_u: 0.5224 - val_loss: 31023.5469 - val_mean_io_u: 0.4794
Epoch 00124: LearningRateScheduler reducing learning rate to 0.00019188471635058244.
Epoch 124/300
105/105 [==] - 439s 4s/step - loss: 15756.4961 - mean_io_u: 0.5373 - val_loss: 35082.0195 - val_mean_io_u: 0.4197
Epoch 00125: LearningRateScheduler reducing learning rate to 0.00019037452710125906.
Epoch 125/300
105/105 [==] - 439s 4s/step - loss: 16814.4590 - mean_io_u: 0.5277 - val_loss: 34495.8633 - val_mean_io_u: 0.4346
Epoch 00126: LearningRateScheduler reducing learning rate to 0.000188859915813123.
Epoch 126/300
105/105 [==] - 439s 4s/step - loss: 17091.5977 - mean_io_u: 0.5220 - val_loss: 33207.6406 - val_mean_io_u: 0.4411
Epoch 00127: LearningRateScheduler reducing learning rate to 0.00018734104858036995.
Epoch 127/300
105/105 [==] - 439s 4s/step - loss: 16243.8096 - mean_io_u: 0.5276 - val_loss: 38811.0781 - val_mean_io_u: 0.4554
Epoch 00128: LearningRateScheduler reducing learning rate to 0.0001858180919639082.
Epoch 128/300
105/105 [==] - 439s 4s/step - loss: 17294.2188 - mean_io_u: 0.5242 - val_loss: 25867.3496 - val_mean_io_u: 0.5094
Epoch 00129: LearningRateScheduler reducing learning rate to 0.00018429121297309283.
Epoch 129/300
105/105 [==] - 439s 4s/step - loss: 15691.8193 - mean_io_u: 0.5382 - val_loss: 33632.3008 - val_mean_io_u: 0.3933
Epoch 00130: LearningRateScheduler reducing learning rate to 0.00018276057904741157.
Epoch 130/300
105/105 [==] - 439s 4s/step - loss: 16711.9922 - mean_io_u: 0.5244 - val_loss: 26973.7695 - val_mean_io_u: 0.5059
Epoch 00131: LearningRateScheduler reducing learning rate to 0.000181226358038123.
Epoch 131/300
105/105 [==] - 439s 4s/step - loss: 15986.9053 - mean_io_u: 0.5332 - val_loss: 25013.0625 - val_mean_io_u: 0.4985
Epoch 00132: LearningRateScheduler reducing learning rate to 0.00017968871818984994.
Epoch 132/300
105/105 [==] - 439s 4s/step - loss: 14841.0586 - mean_io_u: 0.5500 - val_loss: 34063.1523 - val_mean_io_u: 0.3936
Epoch 00133: LearningRateScheduler reducing learning rate to 0.0001781478281221294.
Epoch 133/300
105/105 [==] - 439s 4s/step - loss: 15922.7646 - mean_io_u: 0.5356 - val_loss: 39326.2734 - val_mean_io_u: 0.4607
Epoch 00134: LearningRateScheduler reducing learning rate to 0.00017660385681092153.
Epoch 134/300
105/105 [==] - 439s 4s/step - loss: 15496.8467 - mean_io_u: 0.5408 - val_loss: 30490.5430 - val_mean_io_u: 0.4853
Epoch 00135: LearningRateScheduler reducing learning rate to 0.00017505697357007952.
Epoch 135/300
105/105 [==] - 440s 4s/step - loss: 15590.5410 - mean_io_u: 0.5368 - val_loss: 41027.1172 - val_mean_io_u: 0.3790
Epoch 00136: LearningRateScheduler reducing learning rate to 0.00017350734803278262.
Epoch 136/300
105/105 [==] - 442s 4s/step - loss: 16203.2998 - mean_io_u: 0.5323 - val_loss: 29206.0820 - val_mean_io_u: 0.4849
Epoch 00137: LearningRateScheduler reducing learning rate to 0.00017195515013293362.
Epoch 137/300
105/105 [==] - 439s 4s/step - loss: 15366.5479 - mean_io_u: 0.5419 - val_loss: 43269.9414 - val_mean_io_u: 0.4547
Epoch 00138: LearningRateScheduler reducing learning rate to 0.00017040055008652393.
Epoch 138/300
105/105 [==] - 439s 4s/step - loss: 15838.4717 - mean_io_u: 0.5369 - val_loss: 38508.1680 - val_mean_io_u: 0.4348
Epoch 00139: LearningRateScheduler reducing learning rate to 0.0001688437183729674.
Epoch 139/300
105/105 [==] - 439s 4s/step - loss: 16346.1699 - mean_io_u: 0.5324 - val_loss: 38645.0078 - val_mean_io_u: 0.4444
Epoch 00140: LearningRateScheduler reducing learning rate to 0.00016728482571640535.
Epoch 140/300
105/105 [==] - 439s 4s/step - loss: 16047.3330 - mean_io_u: 0.5365 - val_loss: 42424.5977 - val_mean_io_u: 0.4213
Epoch 00141: LearningRateScheduler reducing learning rate to 0.00016572404306698467.
Epoch 141/300
105/105 [==] - 439s 4s/step - loss: 15129.7031 - mean_io_u: 0.5431 - val_loss: 32691.6191 - val_mean_io_u: 0.4485
Epoch 00142: LearningRateScheduler reducing learning rate to 0.00016416154158211123.
Epoch 142/300
105/105 [==] - 439s 4s/step - loss: 15502.5469 - mean_io_u: 0.5399 - val_loss: 29551.8164 - val_mean_io_u: 0.4484
Epoch 00143: LearningRateScheduler reducing learning rate to 0.0001625974926076807.
Epoch 143/300
105/105 [==] - 439s 4s/step - loss: 15238.3672 - mean_io_u: 0.5422 - val_loss: 32409.0820 - val_mean_io_u: 0.5215
Epoch 00144: LearningRateScheduler reducing learning rate to 0.00016103206765928835.
Epoch 144/300
105/105 [==] - 439s 4s/step - loss: 14863.4492 - mean_io_u: 0.5477 - val_loss: 35020.7578 - val_mean_io_u: 0.4357
Epoch 00145: LearningRateScheduler reducing learning rate to 0.00015946543840342055.
Epoch 145/300
105/105 [==] - 439s 4s/step - loss: 15653.0801 - mean_io_u: 0.5367 - val_loss: 28742.7930 - val_mean_io_u: 0.4910
Epoch 00146: LearningRateScheduler reducing learning rate to 0.00015789777663862943.
Epoch 146/300
105/105 [==] - 439s 4s/step - loss: 15151.9258 - mean_io_u: 0.5452 - val_loss: 30015.6309 - val_mean_io_u: 0.5141
Epoch 00147: LearningRateScheduler reducing learning rate to 0.0001563292542766935.
Epoch 147/300
105/105 [==] - 439s 4s/step - loss: 15686.4990 - mean_io_u: 0.5366 - val_loss: 26746.8066 - val_mean_io_u: 0.5204
Epoch 00148: LearningRateScheduler reducing learning rate to 0.0001547600433237653.
Epoch 148/300
105/105 [==] - 439s 4s/step - loss: 14849.4512 - mean_io_u: 0.5457 - val_loss: 26787.9004 - val_mean_io_u: 0.5005
Epoch 00149: LearningRateScheduler reducing learning rate to 0.00015319031586150938.
Epoch 149/300
105/105 [==] - 439s 4s/step - loss: 15384.6621 - mean_io_u: 0.5408 - val_loss: 42342.4141 - val_mean_io_u: 0.3995
Epoch 00150: LearningRateScheduler reducing learning rate to 0.00015162024402823104.
Epoch 150/300
105/105 [==] - 440s 4s/step - loss: 15260.9531 - mean_io_u: 0.5421 - val_loss: 44724.9922 - val_mean_io_u: 0.4080
Epoch 00151: LearningRateScheduler reducing learning rate to 0.00015005.
Epoch 151/300
105/105 [==] - 440s 4s/step - loss: 15703.0400 - mean_io_u: 0.5371 - val_loss: 42014.1406 - val_mean_io_u: 0.3670
Epoch 00152: LearningRateScheduler reducing learning rate to 0.00014847975597176893.
Epoch 152/300
105/105 [==] - 439s 4s/step - loss: 14999.8779 - mean_io_u: 0.5477 - val_loss: 41360.4258 - val_mean_io_u: 0.4130
Epoch 00153: LearningRateScheduler reducing learning rate to 0.00014690968413849065.
Epoch 153/300
105/105 [==] - 439s 4s/step - loss: 15147.6094 - mean_io_u: 0.5442 - val_loss: 44195.5430 - val_mean_io_u: 0.4053
Epoch 00154: LearningRateScheduler reducing learning rate to 0.00014533995667623465.
Epoch 154/300
105/105 [==] - 439s 4s/step - loss: 15379.9580 - mean_io_u: 0.5374 - val_loss: 38679.0078 - val_mean_io_u: 0.4564
Epoch 00155: LearningRateScheduler reducing learning rate to 0.0001437707457233065.
Epoch 155/300
105/105 [==] - 439s 4s/step - loss: 15133.6992 - mean_io_u: 0.5424 - val_loss: 33237.2891 - val_mean_io_u: 0.4706
Epoch 00156: LearningRateScheduler reducing learning rate to 0.00014220222336137057.
Epoch 156/300
105/105 [==] - 439s 4s/step - loss: 15792.9209 - mean_io_u: 0.5378 - val_loss: 42421.9766 - val_mean_io_u: 0.4122
Epoch 00157: LearningRateScheduler reducing learning rate to 0.00014063456159657948.
Epoch 157/300
105/105 [==] - 439s 4s/step - loss: 14866.0439 - mean_io_u: 0.5469 - val_loss: 35057.2266 - val_mean_io_u: 0.4576
Epoch 00158: LearningRateScheduler reducing learning rate to 0.00013906793234071165.
Epoch 158/300
105/105 [==] - 439s 4s/step - loss: 14899.1494 - mean_io_u: 0.5486 - val_loss: 34308.3750 - val_mean_io_u: 0.4540
Epoch 00159: LearningRateScheduler reducing learning rate to 0.00013750250739231928.
Epoch 159/300
105/105 [==] - 439s 4s/step - loss: 14986.5469 - mean_io_u: 0.5424 - val_loss: 32706.8066 - val_mean_io_u: 0.5045
Epoch 00160: LearningRateScheduler reducing learning rate to 0.00013593845841788877.
Epoch 160/300
105/105 [==] - 441s 4s/step - loss: 15825.5137 - mean_io_u: 0.5364 - val_loss: 28209.9414 - val_mean_io_u: 0.5429
Epoch 00161: LearningRateScheduler reducing learning rate to 0.00013437595693301539.
Epoch 161/300
105/105 [==] - 439s 4s/step - loss: 15648.6895 - mean_io_u: 0.5423 - val_loss: 37491.4844 - val_mean_io_u: 0.4430
Epoch 00162: LearningRateScheduler reducing learning rate to 0.00013281517428359466.
Epoch 162/300
105/105 [==] - 439s 4s/step - loss: 15473.0303 - mean_io_u: 0.5438 - val_loss: 40086.1602 - val_mean_io_u: 0.4003
Epoch 00163: LearningRateScheduler reducing learning rate to 0.00013125628162703257.
Epoch 163/300
105/105 [==] - 439s 4s/step - loss: 14534.6484 - mean_io_u: 0.5505 - val_loss: 37669.7188 - val_mean_io_u: 0.5280
Epoch 00164: LearningRateScheduler reducing learning rate to 0.00012969944991347607.
Epoch 164/300
105/105 [==] - 439s 4s/step - loss: 14084.4209 - mean_io_u: 0.5533 - val_loss: 42714.6250 - val_mean_io_u: 0.4664
Epoch 00165: LearningRateScheduler reducing learning rate to 0.0001281448498670664.
Epoch 165/300
105/105 [==] - 439s 4s/step - loss: 14980.8652 - mean_io_u: 0.5475 - val_loss: 35214.4766 - val_mean_io_u: 0.4721
Epoch 00166: LearningRateScheduler reducing learning rate to 0.0001265926519672174.
Epoch 166/300
105/105 [==] - 439s 4s/step - loss: 15107.9053 - mean_io_u: 0.5475 - val_loss: 40163.2617 - val_mean_io_u: 0.4441
Epoch 00167: LearningRateScheduler reducing learning rate to 0.00012504302642992043.
Epoch 167/300
105/105 [==] - 439s 4s/step - loss: 15310.4600 - mean_io_u: 0.5428 - val_loss: 43188.4688 - val_mean_io_u: 0.4248
Epoch 00168: LearningRateScheduler reducing learning rate to 0.0001234961431890785.
Epoch 168/300
105/105 [==] - 439s 4s/step - loss: 14590.9004 - mean_io_u: 0.5512 - val_loss: 34130.9609 - val_mean_io_u: 0.4800
Epoch 00169: LearningRateScheduler reducing learning rate to 0.00012195217187787059.
Epoch 169/300
105/105 [==] - 439s 4s/step - loss: 15064.8652 - mean_io_u: 0.5298 - val_loss: 48833.5391 - val_mean_io_u: 0.3706
Epoch 00170: LearningRateScheduler reducing learning rate to 0.00012041128181015001.
Epoch 170/300
105/105 [==] - 439s 4s/step - loss: 14781.8125 - mean_io_u: 0.5481 - val_loss: 29504.9023 - val_mean_io_u: 0.4256
Epoch 00171: LearningRateScheduler reducing learning rate to 0.00011887364196187697.
Epoch 171/300
105/105 [==] - 439s 4s/step - loss: 14996.9404 - mean_io_u: 0.5446 - val_loss: 29586.9902 - val_mean_io_u: 0.4676
Epoch 00172: LearningRateScheduler reducing learning rate to 0.00011733942095258844.
Epoch 172/300
105/105 [==] - 439s 4s/step - loss: 15171.8896 - mean_io_u: 0.5439 - val_loss: 35399.5000 - val_mean_io_u: 0.4251
Epoch 00173: LearningRateScheduler reducing learning rate to 0.00011580878702690719.
Epoch 173/300
105/105 [==] - 439s 4s/step - loss: 15654.2832 - mean_io_u: 0.5389 - val_loss: 36140.3047 - val_mean_io_u: 0.4393
Epoch 00174: LearningRateScheduler reducing learning rate to 0.00011428190803609182.
Epoch 174/300
105/105 [==] - 439s 4s/step - loss: 14080.2959 - mean_io_u: 0.5555 - val_loss: 36942.0859 - val_mean_io_u: 0.4264
Epoch 00175: LearningRateScheduler reducing learning rate to 0.00011275895141963004.
Epoch 175/300
105/105 [==] - 439s 4s/step - loss: 13979.7373 - mean_io_u: 0.5575 - val_loss: 25810.1699 - val_mean_io_u: 0.5380
Epoch 00176: LearningRateScheduler reducing learning rate to 0.00011124008418687702.
Epoch 176/300
105/105 [==] - 439s 4s/step - loss: 13951.3096 - mean_io_u: 0.5577 - val_loss: 32986.9883 - val_mean_io_u: 0.4925
Epoch 00177: LearningRateScheduler reducing learning rate to 0.00010972547289874094.
Epoch 177/300
105/105 [==] - 439s 4s/step - loss: 14600.2334 - mean_io_u: 0.5481 - val_loss: 36212.9297 - val_mean_io_u: 0.4245
Epoch 00178: LearningRateScheduler reducing learning rate to 0.00010821528364941755.
Epoch 178/300
105/105 [==] - 439s 4s/step - loss: 15454.8037 - mean_io_u: 0.5417 - val_loss: 27372.0566 - val_mean_io_u: 0.5221
Epoch 00179: LearningRateScheduler reducing learning rate to 0.00010670968204817645.
Epoch 179/300
105/105 [==] - 439s 4s/step - loss: 14609.6143 - mean_io_u: 0.5489 - val_loss: 45445.5781 - val_mean_io_u: 0.4194
Epoch 00180: LearningRateScheduler reducing learning rate to 0.00010520883320119981.
Epoch 180/300
105/105 [==] - 439s 4s/step - loss: 14325.7695 - mean_io_u: 0.5538 - val_loss: 38379.3516 - val_mean_io_u: 0.4185
Epoch 00181: LearningRateScheduler reducing learning rate to 0.00010371290169347664.
Epoch 181/300
105/105 [==] - 439s 4s/step - loss: 15173.7168 - mean_io_u: 0.5472 - val_loss: 38813.4648 - val_mean_io_u: 0.4471
Epoch 00182: LearningRateScheduler reducing learning rate to 0.0001022220515707544.
Epoch 182/300
105/105 [==] - 439s 4s/step - loss: 15060.0059 - mean_io_u: 0.5458 - val_loss: 32711.5840 - val_mean_io_u: 0.4740
Epoch 00183: LearningRateScheduler reducing learning rate to 0.00010073644632154944.
Epoch 183/300
105/105 [==] - 439s 4s/step - loss: 14538.9590 - mean_io_u: 0.5515 - val_loss: 27709.0254 - val_mean_io_u: 0.4558
Epoch 00184: LearningRateScheduler reducing learning rate to 9.925624885921855e-05.
Epoch 184/300
105/105 [==] - 439s 4s/step - loss: 14644.1387 - mean_io_u: 0.5504 - val_loss: 28857.8711 - val_mean_io_u: 0.5074
Epoch 00185: LearningRateScheduler reducing learning rate to 9.778162150409383e-05.
Epoch 185/300
105/105 [==] - 439s 4s/step - loss: 14537.2686 - mean_io_u: 0.5535 - val_loss: 30100.6758 - val_mean_io_u: 0.4568
Epoch 00186: LearningRateScheduler reducing learning rate to 9.631272596568224e-05.
Epoch 186/300
105/105 [==] - 439s 4s/step - loss: 14919.8242 - mean_io_u: 0.5493 - val_loss: 32299.7461 - val_mean_io_u: 0.4748
Epoch 00187: LearningRateScheduler reducing learning rate to 9.484972332493257e-05.
Epoch 187/300
105/105 [==] - 468s 4s/step - loss: 14790.5566 - mean_io_u: 0.5492 - val_loss: 34132.8086 - val_mean_io_u: 0.4599
Epoch 00188: LearningRateScheduler reducing learning rate to 9.339277401657084e-05.
Epoch 188/300
105/105 [==] - 439s 4s/step - loss: 15337.8887 - mean_io_u: 0.5428 - val_loss: 34241.4297 - val_mean_io_u: 0.4788
Epoch 00189: LearningRateScheduler reducing learning rate to 9.194203781150716e-05.
Epoch 189/300
105/105 [==] - 439s 4s/step - loss: 14176.1562 - mean_io_u: 0.5522 - val_loss: 26045.0352 - val_mean_io_u: 0.5178
Epoch 00190: LearningRateScheduler reducing learning rate to 9.049767379931463e-05.
Epoch 190/300
105/105 [==] - 439s 4s/step - loss: 14273.8350 - mean_io_u: 0.5571 - val_loss: 32763.4414 - val_mean_io_u: 0.5322
Epoch 00191: LearningRateScheduler reducing learning rate to 8.905984037078375e-05.
Epoch 191/300
105/105 [==] - 439s 4s/step - loss: 14640.9912 - mean_io_u: 0.5338 - val_loss: 40018.9766 - val_mean_io_u: 0.4519
Epoch 00192: LearningRateScheduler reducing learning rate to 8.762869520055285e-05.
Epoch 192/300
105/105 [==] - 439s 4s/step - loss: 14759.4453 - mean_io_u: 0.5493 - val_loss: 33715.3281 - val_mean_io_u: 0.4344
Epoch 00193: LearningRateScheduler reducing learning rate to 8.620439522981734e-05.
Epoch 193/300
105/105 [==] - 439s 4s/step - loss: 13657.6465 - mean_io_u: 0.5630 - val_loss: 31174.0781 - val_mean_io_u: 0.4388
Epoch 00194: LearningRateScheduler reducing learning rate to 8.478709664911951e-05.
Epoch 194/300
105/105 [==] - 439s 4s/step - loss: 14163.4580 - mean_io_u: 0.5557 - val_loss: 30409.0059 - val_mean_io_u: 0.4512
Epoch 00195: LearningRateScheduler reducing learning rate to 8.337695488122015e-05.
Epoch 195/300
105/105 [==] - 439s 4s/step - loss: 13633.5303 - mean_io_u: 0.5604 - val_loss: 30746.8555 - val_mean_io_u: 0.4689
Epoch 00196: LearningRateScheduler reducing learning rate to 8.197412456405496e-05.
Epoch 196/300
105/105 [==] - 439s 4s/step - loss: 14424.0098 - mean_io_u: 0.5509 - val_loss: 37804.8555 - val_mean_io_u: 0.4550
Epoch 00197: LearningRateScheduler reducing learning rate to 8.057875953377676e-05.
Epoch 197/300
105/105 [==] - 439s 4s/step - loss: 13904.6211 - mean_io_u: 0.5537 - val_loss: 26515.2500 - val_mean_io_u: 0.5394
Epoch 00198: LearningRateScheduler reducing learning rate to 7.919101280788536e-05.
Epoch 198/300
105/105 [==] - 439s 4s/step - loss: 14204.1436 - mean_io_u: 0.5584 - val_loss: 30020.1211 - val_mean_io_u: 0.5011
Epoch 00199: LearningRateScheduler reducing learning rate to 7.781103656844776e-05.
Epoch 199/300
105/105 [==] - 439s 4s/step - loss: 14643.4521 - mean_io_u: 0.5498 - val_loss: 31174.7871 - val_mean_io_u: 0.4424
Epoch 00200: LearningRateScheduler reducing learning rate to 7.643898214540963e-05.
Epoch 200/300
105/105 [==] - 439s 4s/step - loss: 14203.6465 - mean_io_u: 0.5557 - val_loss: 30767.7656 - val_mean_io_u: 0.5028
Epoch 00201: LearningRateScheduler reducing learning rate to 7.507499999999997e-05.
Epoch 201/300
105/105 [==] - 439s 4s/step - loss: 13700.2197 - mean_io_u: 0.5625 - val_loss: 41935.9883 - val_mean_io_u: 0.4834
Epoch 00202: LearningRateScheduler reducing learning rate to 7.371923970823181e-05.
Epoch 202/300
105/105 [==] - 439s 4s/step - loss: 14243.2266 - mean_io_u: 0.5527 - val_loss: 45971.5117 - val_mean_io_u: 0.3965
Epoch 00203: LearningRateScheduler reducing learning rate to 7.237184994449914e-05.
Epoch 203/300
105/105 [==] - 439s 4s/step - loss: 14570.1348 - mean_io_u: 0.5486 - val_loss: 36789.5000 - val_mean_io_u: 0.4306
Epoch 00204: LearningRateScheduler reducing learning rate to 7.10329784652732e-05.
Epoch 204/300
105/105 [==] - 439s 4s/step - loss: 13828.4834 - mean_io_u: 0.5593 - val_loss: 42057.9688 - val_mean_io_u: 0.4318
Epoch 00205: LearningRateScheduler reducing learning rate to 6.970277209289948e-05.
Epoch 205/300
105/105 [==] - 439s 4s/step - loss: 14754.8467 - mean_io_u: 0.5485 - val_loss: 39776.7734 - val_mean_io_u: 0.3951
Epoch 00206: LearningRateScheduler reducing learning rate to 6.838137669949673e-05.
Epoch 206/300
105/105 [==] - 439s 4s/step - loss: 13831.8857 - mean_io_u: 0.5552 - val_loss: 41408.0000 - val_mean_io_u: 0.4243
Epoch 00207: LearningRateScheduler reducing learning rate to 6.706893719096057e-05.
Epoch 207/300
105/105 [==] - 439s 4s/step - loss: 14524.7061 - mean_io_u: 0.5532 - val_loss: 48952.7969 - val_mean_io_u: 0.3976
Epoch 00208: LearningRateScheduler reducing learning rate to 6.576559749107305e-05.
Epoch 208/300
105/105 [==] - 439s 4s/step - loss: 14147.7705 - mean_io_u: 0.5547 - val_loss: 33560.4258 - val_mean_io_u: 0.4895
Epoch 00209: LearningRateScheduler reducing learning rate to 6.447150052571948e-05.
Epoch 209/300
105/105 [==] - 439s 4s/step - loss: 14234.3896 - mean_io_u: 0.5516 - val_loss: 29967.4727 - val_mean_io_u: 0.4647
Epoch 00210: LearningRateScheduler reducing learning rate to 6.318678820721529e-05.
Epoch 210/300
105/105 [==] - 439s 4s/step - loss: 13664.8105 - mean_io_u: 0.5649 - val_loss: 38095.2734 - val_mean_io_u: 0.4894
Epoch 00211: LearningRateScheduler reducing learning rate to 6.191160141874366e-05.
Epoch 211/300
105/105 [==] - 439s 4s/step - loss: 14412.4248 - mean_io_u: 0.5549 - val_loss: 33475.1875 - val_mean_io_u: 0.4442
Epoch 00212: LearningRateScheduler reducing learning rate to 6.0646079998905885e-05.
Epoch 212/300
105/105 [==] - 439s 4s/step - loss: 14136.7988 - mean_io_u: 0.5590 - val_loss: 32640.9531 - val_mean_io_u: 0.4696
Epoch 00213: LearningRateScheduler reducing learning rate to 5.939036272638689e-05.
Epoch 213/300
105/105 [==] - 440s 4s/step - loss: 14590.7520 - mean_io_u: 0.5527 - val_loss: 34158.7539 - val_mean_io_u: 0.5053
Epoch 00214: LearningRateScheduler reducing learning rate to 5.8144587304736207e-05.
Epoch 214/300
105/105 [==] - 439s 4s/step - loss: 14252.8418 - mean_io_u: 0.5518 - val_loss: 42583.5117 - val_mean_io_u: 0.4413
Epoch 00215: LearningRateScheduler reducing learning rate to 5.6908890347267367e-05.
Epoch 215/300
105/105 [==] - 439s 4s/step - loss: 14509.8428 - mean_io_u: 0.5488 - val_loss: 39850.9844 - val_mean_io_u: 0.4784
Epoch 00216: LearningRateScheduler reducing learning rate to 5.56834073620769e-05.
Epoch 216/300
105/105 [==] - 439s 4s/step - loss: 14100.8154 - mean_io_u: 0.5572 - val_loss: 34432.9297 - val_mean_io_u: 0.4208
Epoch 00217: LearningRateScheduler reducing learning rate to 5.446827273718397e-05.
Epoch 217/300
105/105 [==] - 439s 4s/step - loss: 13779.1221 - mean_io_u: 0.5590 - val_loss: 33797.9766 - val_mean_io_u: 0.4345
Epoch 00218: LearningRateScheduler reducing learning rate to 5.326361972579363e-05.
Epoch 218/300
105/105 [==] - 439s 4s/step - loss: 14427.8564 - mean_io_u: 0.5499 - val_loss: 38356.6836 - val_mean_io_u: 0.3999
Epoch 00219: LearningRateScheduler reducing learning rate to 5.206958043168373e-05.
Epoch 219/300
105/105 [==] - 439s 4s/step - loss: 14423.2373 - mean_io_u: 0.5496 - val_loss: 34448.3672 - val_mean_io_u: 0.4738
Epoch 00220: LearningRateScheduler reducing learning rate to 5.088628579471837e-05.
Epoch 220/300
105/105 [==] - 439s 4s/step - loss: 13499.8672 - mean_io_u: 0.5649 - val_loss: 40659.5977 - val_mean_io_u: 0.4423
Epoch 00221: LearningRateScheduler reducing learning rate to 4.9713865576489204e-05.
Epoch 221/300
105/105 [==] - 439s 4s/step - loss: 13214.4062 - mean_io_u: 0.5645 - val_loss: 37023.3906 - val_mean_io_u: 0.4415
Epoch 00222: LearningRateScheduler reducing learning rate to 4.8552448346085036e-05.
Epoch 222/300
105/105 [==] - 439s 4s/step - loss: 13621.0322 - mean_io_u: 0.5641 - val_loss: 39336.8359 - val_mean_io_u: 0.5182
Epoch 00223: LearningRateScheduler reducing learning rate to 4.740216146599312e-05.
Epoch 223/300
105/105 [==] - 439s 4s/step - loss: 14423.4717 - mean_io_u: 0.5503 - val_loss: 36699.8633 - val_mean_io_u: 0.4428
Epoch 00224: LearningRateScheduler reducing learning rate to 4.6263131078132514e-05.
Epoch 224/300
105/105 [==] - 439s 4s/step - loss: 14319.3672 - mean_io_u: 0.5513 - val_loss: 31253.1133 - val_mean_io_u: 0.4316
Epoch 00225: LearningRateScheduler reducing learning rate to 4.513548209002085e-05.
Epoch 225/300
105/105 [==] - 439s 4s/step - loss: 13973.7246 - mean_io_u: 0.5550 - val_loss: 40381.6328 - val_mean_io_u: 0.3943
Epoch 00226: LearningRateScheduler reducing learning rate to 4.4019338161077204e-05.
Epoch 226/300
105/105 [==] - 439s 4s/step - loss: 13087.0059 - mean_io_u: 0.5683 - val_loss: 38013.1484 - val_mean_io_u: 0.4428
Epoch 00227: LearningRateScheduler reducing learning rate to 4.291482168906117e-05.
Epoch 227/300
105/105 [==] - 439s 4s/step - loss: 14050.5684 - mean_io_u: 0.5552 - val_loss: 41338.3867 - val_mean_io_u: 0.4950
Epoch 00228: LearningRateScheduler reducing learning rate to 4.182205379665059e-05.
Epoch 228/300
105/105 [==] - 439s 4s/step - loss: 13758.3418 - mean_io_u: 0.5599 - val_loss: 38963.6328 - val_mean_io_u: 0.4688
Epoch 00229: LearningRateScheduler reducing learning rate to 4.074115431815937e-05.
Epoch 229/300
105/105 [==] - 439s 4s/step - loss: 13745.5801 - mean_io_u: 0.5608 - val_loss: 53638.7617 - val_mean_io_u: 0.3923
Epoch 00230: LearningRateScheduler reducing learning rate to 3.9672241786395886e-05.
Epoch 230/300
105/105 [==] - 439s 4s/step - loss: 13849.7422 - mean_io_u: 0.5598 - val_loss: 36536.1836 - val_mean_io_u: 0.4519
Epoch 00231: LearningRateScheduler reducing learning rate to 3.8615433419664717e-05.
Epoch 231/300
105/105 [==] - 439s 4s/step - loss: 13429.1416 - mean_io_u: 0.5643 - val_loss: 39454.6641 - val_mean_io_u: 0.4574
Epoch 00232: LearningRateScheduler reducing learning rate to 3.75708451089126e-05.
Epoch 232/300
105/105 [==] - 439s 4s/step - loss: 14164.0566 - mean_io_u: 0.5528 - val_loss: 38913.8359 - val_mean_io_u: 0.4487
Epoch 00233: LearningRateScheduler reducing learning rate to 3.653859140501914e-05.
Epoch 233/300
105/105 [==] - 439s 4s/step - loss: 13199.7168 - mean_io_u: 0.5677 - val_loss: 43196.8789 - val_mean_io_u: 0.4751
Epoch 00234: LearningRateScheduler reducing learning rate to 3.551878550623541e-05.
Epoch 234/300
105/105 [==] - 439s 4s/step - loss: 14364.4043 - mean_io_u: 0.5510 - val_loss: 48101.1953 - val_mean_io_u: 0.4453
Epoch 00235: LearningRateScheduler reducing learning rate to 3.4511539245770414e-05.
Epoch 235/300
105/105 [==] - 439s 4s/step - loss: 13883.1836 - mean_io_u: 0.5573 - val_loss: 38235.3203 - val_mean_io_u: 0.4665
Epoch 00236: LearningRateScheduler reducing learning rate to 3.3516963079527215e-05.
Epoch 236/300
105/105 [==] - 439s 4s/step - loss: 13697.2041 - mean_io_u: 0.5609 - val_loss: 29029.4883 - val_mean_io_u: 0.5325
Epoch 00237: LearningRateScheduler reducing learning rate to 3.2535166073990334e-05.
Epoch 237/300
105/105 [==] - 439s 4s/step - loss: 13871.2295 - mean_io_u: 0.5575 - val_loss: 42103.8438 - val_mean_io_u: 0.4221
Epoch 00238: LearningRateScheduler reducing learning rate to 3.156625589426525e-05.
Epoch 238/300
105/105 [==] - 439s 4s/step - loss: 14293.4121 - mean_io_u: 0.5559 - val_loss: 49225.9609 - val_mean_io_u: 0.3955
Epoch 00239: LearningRateScheduler reducing learning rate to 3.0610338792271784e-05.
その6(ResNetをさらに少し値をかえる。)
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet152
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 4
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.2
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 5.0
Zoom --> 0.1
Channel Shift --> 0.1
Epoch 00282: LearningRateScheduler reducing learning rate to 3.0583373654728226e-06.
Epoch 282/300
105/105 [==] - 336s 3s/step - loss: 9642.9912 - mean_io_u: 0.6224 - val_loss: 42120.7891 - val_mean_io_u: 0.4211
Epoch 00283: LearningRateScheduler reducing learning rate to 2.7560267532331423e-06.
Epoch 283/300
105/105 [==] - 343s 3s/step - loss: 9774.2246 - mean_io_u: 0.6220 - val_loss: 39625.9453 - val_mean_io_u: 0.4381
Epoch 00284: LearningRateScheduler reducing learning rate to 2.4698685850121796e-06.
Epoch 284/300
105/105 [==] - 347s 3s/step - loss: 9940.4307 - mean_io_u: 0.6170 - val_loss: 38310.8945 - val_mean_io_u: 0.5161
Epoch 00285: LearningRateScheduler reducing learning rate to 2.1998942412777778e-06.
Epoch 285/300
105/105 [==] - 347s 3s/step - loss: 9922.2578 - mean_io_u: 0.6183 - val_loss: 38778.2109 - val_mean_io_u: 0.4697
Epoch 00286: LearningRateScheduler reducing learning rate to 1.9461333277591077e-06.
Epoch 286/300
105/105 [==] - 348s 3s/step - loss: 10017.0273 - mean_io_u: 0.6204 - val_loss: 33931.5859 - val_mean_io_u: 0.4451
Epoch 00287: LearningRateScheduler reducing learning rate to 1.7086136721998994e-06.
Epoch 287/300
105/105 [==] - 367s 3s/step - loss: 10207.9473 - mean_io_u: 0.6153 - val_loss: 36609.2266 - val_mean_io_u: 0.4528
Epoch 00288: LearningRateScheduler reducing learning rate to 1.4873613213070326e-06.
Epoch 288/300
105/105 [==] - 338s 3s/step - loss: 9705.6338 - mean_io_u: 0.6212 - val_loss: 39513.2461 - val_mean_io_u: 0.4077
Epoch 00289: LearningRateScheduler reducing learning rate to 1.2824005378940589e-06.
Epoch 289/300
105/105 [==] - 340s 3s/step - loss: 9609.6250 - mean_io_u: 0.6251 - val_loss: 35499.0898 - val_mean_io_u: 0.4380
Epoch 00290: LearningRateScheduler reducing learning rate to 1.093753798220553e-06.
Epoch 290/300
105/105 [==] - 332s 3s/step - loss: 10076.4199 - mean_io_u: 0.6170 - val_loss: 42613.4648 - val_mean_io_u: 0.4285
Epoch 00291: LearningRateScheduler reducing learning rate to 9.214417895274202e-07.
Epoch 291/300
105/105 [==] - 337s 3s/step - loss: 9577.9072 - mean_io_u: 0.6227 - val_loss: 33922.4375 - val_mean_io_u: 0.4640
Epoch 00292: LearningRateScheduler reducing learning rate to 7.654834077681545e-07.
Epoch 292/300
105/105 [==] - 354s 3s/step - loss: 9645.2383 - mean_io_u: 0.6223 - val_loss: 35155.5859 - val_mean_io_u: 0.4655
Epoch 00293: LearningRateScheduler reducing learning rate to 6.258957555368196e-07.
Epoch 293/300
105/105 [==] - 338s 3s/step - loss: 9825.3242 - mean_io_u: 0.6220 - val_loss: 38539.3555 - val_mean_io_u: 0.4084
Epoch 00294: LearningRateScheduler reducing learning rate to 5.026941401925086e-07.
Epoch 294/300
105/105 [==] - 340s 3s/step - loss: 10266.7412 - mean_io_u: 0.6160 - val_loss: 42251.1406 - val_mean_io_u: 0.3362
Epoch 00295: LearningRateScheduler reducing learning rate to 3.958920721806797e-07.
Epoch 295/300
105/105 [==] - 355s 3s/step - loss: 9449.6855 - mean_io_u: 0.6282 - val_loss: 46164.1172 - val_mean_io_u: 0.4545
Epoch 00296: LearningRateScheduler reducing learning rate to 3.055012635516537e-07.
Epoch 296/300
105/105 [==] - 335s 3s/step - loss: 9814.1074 - mean_io_u: 0.6232 - val_loss: 39850.5703 - val_mean_io_u: 0.4511
Epoch 00297: LearningRateScheduler reducing learning rate to 2.3153162667618543e-07.
Epoch 297/300
105/105 [==] - 377s 4s/step - loss: 9784.8115 - mean_io_u: 0.6213 - val_loss: 32465.5430 - val_mean_io_u: 0.4685
Epoch 00298: LearningRateScheduler reducing learning rate to 1.7399127315854657e-07.
Epoch 298/300
105/105 [==] - 348s 3s/step - loss: 10461.2939 - mean_io_u: 0.6065 - val_loss: 42474.8828 - val_mean_io_u: 0.4353
Epoch 00299: LearningRateScheduler reducing learning rate to 1.3288651294691732e-07.
Epoch 299/300
105/105 [==] - 344s 3s/step - loss: 9802.5967 - mean_io_u: 0.6220 - val_loss: 26617.5840 - val_mean_io_u: 0.5529
Epoch 00300: LearningRateScheduler reducing learning rate to 1.0822185364145413e-07.
Epoch 300/300
105/105 [==] - 337s 3s/step - loss: 10342.7090 - mean_io_u: 0.6079 - val_loss: 38630.7422 - val_mean_io_u: 0.4652
その7(その前のに対して、オーグメンテーションをかえる)
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> DeepLabV3Plus
Base Model --> ResNet152
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 4
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 15.0
Zoom --> 0.1
Channel Shift --> 0.1
Epoch 00188: LearningRateScheduler reducing learning rate to 9.339277401657084e-05.
Epoch 188/300
105/105 [==] - 527s 5s/step - loss: 15350.4150 - mean_io_u: 0.5448 - val_loss: 25813.9512 - val_mean_io_u: 0.4830
Epoch 00189: LearningRateScheduler reducing learning rate to 9.194203781150716e-05.
Epoch 189/300
105/105 [==] - 527s 5s/step - loss: 14895.3438 - mean_io_u: 0.5475 - val_loss: 35445.7461 - val_mean_io_u: 0.4441
Epoch 00190: LearningRateScheduler reducing learning rate to 9.049767379931463e-05.
Epoch 190/300
105/105 [==] - 527s 5s/step - loss: 14428.6836 - mean_io_u: 0.5522 - val_loss: 51286.9258 - val_mean_io_u: 0.3764
Epoch 00191: LearningRateScheduler reducing learning rate to 8.905984037078375e-05.
Epoch 191/300
105/105 [==] - 527s 5s/step - loss: 15003.8516 - mean_io_u: 0.5479 - val_loss: 42903.3438 - val_mean_io_u: 0.4295
Epoch 00192: LearningRateScheduler reducing learning rate to 8.762869520055285e-05.
Epoch 192/300
105/105 [==] - 527s 5s/step - loss: 15446.7051 - mean_io_u: 0.5407 - val_loss: 37077.4375 - val_mean_io_u: 0.4162
Epoch 00193: LearningRateScheduler reducing learning rate to 8.620439522981734e-05.
Epoch 193/300
105/105 [==] - 527s 5s/step - loss: 15499.0967 - mean_io_u: 0.5430 - val_loss: 27281.6191 - val_mean_io_u: 0.4473
Epoch 00194: LearningRateScheduler reducing learning rate to 8.478709664911951e-05.
Epoch 194/300
105/105 [==] - 527s 5s/step - loss: 15826.6797 - mean_io_u: 0.5339 - val_loss: 31886.1836 - val_mean_io_u: 0.4124
Epoch 00195: LearningRateScheduler reducing learning rate to 8.337695488122015e-05.
Epoch 195/300
105/105 [==] - 527s 5s/step - loss: 15473.5068 - mean_io_u: 0.5385 - val_loss: 40918.0039 - val_mean_io_u: 0.4251
Epoch 00196: LearningRateScheduler reducing learning rate to 8.197412456405496e-05.
Epoch 196/300
105/105 [==] - 528s 5s/step - loss: 15493.5537 - mean_io_u: 0.5425 - val_loss: 33786.1367 - val_mean_io_u: 0.4137
Epoch 00197: LearningRateScheduler reducing learning rate to 8.057875953377676e-05.
Epoch 197/300
105/105 [==] - 527s 5s/step - loss: 15747.3652 - mean_io_u: 0.5360 - val_loss: 34643.0234 - val_mean_io_u: 0.4354
Epoch 00198: LearningRateScheduler reducing learning rate to 7.919101280788536e-05.
Epoch 198/300
105/105 [==] - 527s 5s/step - loss: 15162.3525 - mean_io_u: 0.5432 - val_loss: 47692.2109 - val_mean_io_u: 0.4041
Epoch 00199: LearningRateScheduler reducing learning rate to 7.781103656844776e-05.
Epoch 199/300
105/105 [==] - 527s 5s/step - loss: 15366.5439 - mean_io_u: 0.5403 - val_loss: 37444.9453 - val_mean_io_u: 0.4167
Epoch 00200: LearningRateScheduler reducing learning rate to 7.643898214540963e-05.
Epoch 200/300
105/105 [==] - 527s 5s/step - loss: 15176.5488 - mean_io_u: 0.5426 - val_loss: 30566.3789 - val_mean_io_u: 0.4977
Epoch 00201: LearningRateScheduler reducing learning rate to 7.507499999999997e-05.
Epoch 201/300
105/105 [==] - 527s 5s/step - loss: 15491.7168 - mean_io_u: 0.5437 - val_loss: 43589.2109 - val_mean_io_u: 0.4045
Epoch 00202: LearningRateScheduler reducing learning rate to 7.371923970823181e-05.
Epoch 202/300
105/105 [==] - 527s 5s/step - loss: 15268.3857 - mean_io_u: 0.5424 - val_loss: 36250.5547 - val_mean_io_u: 0.4588
Epoch 00203: LearningRateScheduler reducing learning rate to 7.237184994449914e-05.
Epoch 203/300
105/105 [==] - 527s 5s/step - loss: 15150.2100 - mean_io_u: 0.5467 - val_loss: 25390.1094 - val_mean_io_u: 0.4964
Epoch 00204: LearningRateScheduler reducing learning rate to 7.10329784652732e-05.
Epoch 204/300
105/105 [==] - 527s 5s/step - loss: 14917.7900 - mean_io_u: 0.5478 - val_loss: 42530.0859 - val_mean_io_u: 0.3851
Epoch 00205: LearningRateScheduler reducing learning rate to 6.970277209289948e-05.
Epoch 205/300
105/105 [==] - 527s 5s/step - loss: 15090.8779 - mean_io_u: 0.5463 - val_loss: 32042.3184 - val_mean_io_u: 0.5052
Epoch 00206: LearningRateScheduler reducing learning rate to 6.838137669949673e-05.
Epoch 206/300
105/105 [==] - 527s 5s/step - loss: 14266.7305 - mean_io_u: 0.5533 - val_loss: 44365.8516 - val_mean_io_u: 0.4154
Epoch 00207: LearningRateScheduler reducing learning rate to 6.706893719096057e-05.
Epoch 207/300
105/105 [==] - 527s 5s/step - loss: 15473.3535 - mean_io_u: 0.5420 - val_loss: 40047.4453 - val_mean_io_u: 0.4887
Epoch 00208: LearningRateScheduler reducing learning rate to 6.576559749107305e-05.
Epoch 208/300
105/105 [==] - 527s 5s/step - loss: 14821.1270 - mean_io_u: 0.5519 - val_loss: 38655.5898 - val_mean_io_u: 0.4361
あまり、興味深い結果ではない。
その8(VGG19へ)
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> UNet
Base Model --> VGG19
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 4
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 15.0
Zoom --> 0.1
Channel Shift --> 0.1
Epoch 00278: LearningRateScheduler reducing learning rate to 4.4284147275500865e-06.
Epoch 278/300
105/105 [==] - 262s 2s/step - loss: 13041.4004 - mean_io_u: 0.5759 - val_loss: 36836.8672 - val_mean_io_u: 0.4096
Epoch 00279: LearningRateScheduler reducing learning rate to 4.061843514169614e-06.
Epoch 279/300
105/105 [==] - 262s 2s/step - loss: 13144.2178 - mean_io_u: 0.5734 - val_loss: 30244.1719 - val_mean_io_u: 0.4554
Epoch 00280: LearningRateScheduler reducing learning rate to 3.7112815472848386e-06.
Epoch 280/300
105/105 [==] - 262s 2s/step - loss: 12966.4160 - mean_io_u: 0.5748 - val_loss: 33730.5156 - val_mean_io_u: 0.4099
Epoch 00281: LearningRateScheduler reducing learning rate to 3.3767672699658532e-06.
Epoch 281/300
105/105 [==] - 262s 2s/step - loss: 12453.0010 - mean_io_u: 0.5844 - val_loss: 36161.7305 - val_mean_io_u: 0.4274
Epoch 00282: LearningRateScheduler reducing learning rate to 3.0583373654728226e-06.
Epoch 282/300
105/105 [==] - 262s 2s/step - loss: 12499.7373 - mean_io_u: 0.5800 - val_loss: 18748.4023 - val_mean_io_u: 0.5515
Epoch 00283: LearningRateScheduler reducing learning rate to 2.7560267532331423e-06.
Epoch 283/300
105/105 [==] - 262s 2s/step - loss: 13150.8936 - mean_io_u: 0.5765 - val_loss: 25297.3223 - val_mean_io_u: 0.5206
Epoch 00284: LearningRateScheduler reducing learning rate to 2.4698685850121796e-06.
Epoch 284/300
105/105 [==] - 262s 2s/step - loss: 13421.9834 - mean_io_u: 0.5681 - val_loss: 28623.0254 - val_mean_io_u: 0.5018
Epoch 00285: LearningRateScheduler reducing learning rate to 2.1998942412777778e-06.
Epoch 285/300
105/105 [==] - 262s 2s/step - loss: 13142.8691 - mean_io_u: 0.5742 - val_loss: 32247.8555 - val_mean_io_u: 0.4657
Epoch 00286: LearningRateScheduler reducing learning rate to 1.9461333277591077e-06.
Epoch 286/300
105/105 [==] - 262s 2s/step - loss: 12577.5742 - mean_io_u: 0.5820 - val_loss: 33489.8945 - val_mean_io_u: 0.4923
Epoch 00287: LearningRateScheduler reducing learning rate to 1.7086136721998994e-06.
Epoch 287/300
105/105 [==] - 262s 2s/step - loss: 13352.0547 - mean_io_u: 0.5728 - val_loss: 26837.2656 - val_mean_io_u: 0.4971
Epoch 00288: LearningRateScheduler reducing learning rate to 1.4873613213070326e-06.
Epoch 288/300
105/105 [==] - 262s 2s/step - loss: 12696.1904 - mean_io_u: 0.5782 - val_loss: 26888.3340 - val_mean_io_u: 0.4703
Epoch 00289: LearningRateScheduler reducing learning rate to 1.2824005378940589e-06.
Epoch 289/300
105/105 [==] - 262s 2s/step - loss: 13743.4229 - mean_io_u: 0.5652 - val_loss: 25023.8652 - val_mean_io_u: 0.5094
Epoch 00290: LearningRateScheduler reducing learning rate to 1.093753798220553e-06.
Epoch 290/300
105/105 [==] - 262s 2s/step - loss: 13030.3672 - mean_io_u: 0.5754 - val_loss: 29277.4004 - val_mean_io_u: 0.4536
Epoch 00291: LearningRateScheduler reducing learning rate to 9.214417895274202e-07.
Epoch 291/300
105/105 [==] - 262s 2s/step - loss: 13075.4102 - mean_io_u: 0.5749 - val_loss: 28388.0273 - val_mean_io_u: 0.4484
Epoch 00292: LearningRateScheduler reducing learning rate to 7.654834077681545e-07.
Epoch 292/300
105/105 [==] - 262s 2s/step - loss: 12329.8584 - mean_io_u: 0.5826 - val_loss: 28064.2461 - val_mean_io_u: 0.5163
Epoch 00293: LearningRateScheduler reducing learning rate to 6.258957555368196e-07.
Epoch 293/300
105/105 [==] - 262s 2s/step - loss: 12244.7539 - mean_io_u: 0.5866 - val_loss: 27515.9336 - val_mean_io_u: 0.5570
Epoch 00294: LearningRateScheduler reducing learning rate to 5.026941401925086e-07.
Epoch 294/300
105/105 [==] - 262s 2s/step - loss: 12972.5166 - mean_io_u: 0.5735 - val_loss: 27562.1309 - val_mean_io_u: 0.4888
Epoch 00295: LearningRateScheduler reducing learning rate to 3.958920721806797e-07.
Epoch 295/300
105/105 [==] - 262s 2s/step - loss: 12098.0410 - mean_io_u: 0.5870 - val_loss: 30404.6992 - val_mean_io_u: 0.4871
Epoch 00296: LearningRateScheduler reducing learning rate to 3.055012635516537e-07.
Epoch 296/300
105/105 [==] - 262s 2s/step - loss: 13095.4434 - mean_io_u: 0.5764 - val_loss: 30395.1992 - val_mean_io_u: 0.5303
Epoch 00297: LearningRateScheduler reducing learning rate to 2.3153162667618543e-07.
Epoch 297/300
105/105 [==] - 262s 2s/step - loss: 13845.8545 - mean_io_u: 0.5634 - val_loss: 26944.0059 - val_mean_io_u: 0.4717
Epoch 00298: LearningRateScheduler reducing learning rate to 1.7399127315854657e-07.
Epoch 298/300
105/105 [==] - 262s 2s/step - loss: 13043.8018 - mean_io_u: 0.5749 - val_loss: 34509.7266 - val_mean_io_u: 0.4517
Epoch 00299: LearningRateScheduler reducing learning rate to 1.3288651294691732e-07.
Epoch 299/300
105/105 [==] - 262s 2s/step - loss: 13175.7002 - mean_io_u: 0.5716 - val_loss: 29966.4961 - val_mean_io_u: 0.4763
Epoch 00300: LearningRateScheduler reducing learning rate to 1.0822185364145413e-07.
Epoch 300/300
105/105 [==] - 262s 2s/step - loss: 13899.7666 - mean_io_u: 0.5674 - val_loss: 30677.0117 - val_mean_io_u: 0.5009
その9(VGG16へ)
***** Begin training *****
Dataset --> CamVid
Num Images --> 421
Model --> UNet
Base Model --> VGG16
Crop Height --> 256
Crop Width --> 256
Num Epochs --> 300
Initial Epoch --> 0
Batch Size --> 4
Num Classes --> 32
Data Augmentation:
Data Augmentation Rate --> 0.5
Vertical Flip --> False
Horizontal Flip --> True
Brightness Alteration --> None
Rotation --> 15.0
Zoom --> 0.1
Channel Shift --> 0.1
Epoch 79/300
105/105 [==] - 231s 2s/step - loss: 21093.9590 - mean_io_u: 0.4881 - val_loss: 24889.8906 - val_mean_io_u: 0.4581
Epoch 00080: LearningRateScheduler reducing learning rate to 0.00025154755165391495.
Epoch 80/300
105/105 [==] - 231s 2s/step - loss: 21166.4648 - mean_io_u: 0.4843 - val_loss: 27520.2227 - val_mean_io_u: 0.4448
Epoch 00081: LearningRateScheduler reducing learning rate to 0.00025038613442351075.
Epoch 81/300
105/105 [==] - 231s 2s/step - loss: 20243.4414 - mean_io_u: 0.4927 - val_loss: 23198.8301 - val_mean_io_u: 0.5450
Epoch 00082: LearningRateScheduler reducing learning rate to 0.0002492137142052816.
Epoch 82/300
105/105 [==] - 231s 2s/step - loss: 21469.2344 - mean_io_u: 0.4804 - val_loss: 28869.9902 - val_mean_io_u: 0.4203
Epoch 00083: LearningRateScheduler reducing learning rate to 0.00024803041956831634.
Epoch 83/300
105/105 [==] - 231s 2s/step - loss: 19920.8145 - mean_io_u: 0.4965 - val_loss: 30445.1348 - val_mean_io_u: 0.3874
Epoch 00084: LearningRateScheduler reducing learning rate to 0.0002468363802742064.
Epoch 84/300
105/105 [==] - 231s 2s/step - loss: 19678.6504 - mean_io_u: 0.5019 - val_loss: 32337.4023 - val_mean_io_u: 0.4333
Epoch 00085: LearningRateScheduler reducing learning rate to 0.000245631727262816.
Epoch 85/300
105/105 [==] - 231s 2s/step - loss: 20097.7812 - mean_io_u: 0.4944 - val_loss: 29057.7617 - val_mean_io_u: 0.4194
Epoch 00086: LearningRateScheduler reducing learning rate to 0.0002444165926379231.
Epoch 86/300
105/105 [==] - 231s 2s/step - loss: 20126.7305 - mean_io_u: 0.4947 - val_loss: 28013.0059 - val_mean_io_u: 0.4356
Epoch 00087: LearningRateScheduler reducing learning rate to 0.00024319110965273264.
Epoch 87/300
105/105 [==] - 231s 2s/step - loss: 21490.9336 - mean_io_u: 0.4830 - val_loss: 43573.7695 - val_mean_io_u: 0.3367
Epoch 00088: LearningRateScheduler reducing learning rate to 0.0002419554126952638.
Epoch 88/300
105/105 [==] - 231s 2s/step - loss: 20464.9824 - mean_io_u: 0.4899 - val_loss: 30465.7402 - val_mean_io_u: 0.4155
Epoch 00089: LearningRateScheduler reducing learning rate to 0.00024070963727361311.
Epoch 89/300
105/105 [==] - 231s 2s/step - loss: 19254.0059 - mean_io_u: 0.5030 - val_loss: 34606.4609 - val_mean_io_u: 0.4253
Epoch 00090: LearningRateScheduler reducing learning rate to 0.00023945392000109406.
Epoch 90/300
105/105 [==] - 231s 2s/step - loss: 19508.3613 - mean_io_u: 0.5028 - val_loss: 23513.0078 - val_mean_io_u: 0.4602
Epoch 00091: LearningRateScheduler reducing learning rate to 0.00023818839858125631.
Epoch 91/300
105/105 [==] - 231s 2s/step - loss: 20069.9258 - mean_io_u: 0.4982 - val_loss: 29752.0742 - val_mean_io_u: 0.3745
Epoch 00092: LearningRateScheduler reducing learning rate to 0.00023691321179278464.
Epoch 92/300
105/105 [==] - 231s 2s/step - loss: 18906.5195 - mean_io_u: 0.5077 - val_loss: 25509.8203 - val_mean_io_u: 0.4449
まとめ
特になし。
まだ、記載はじめなので。。
グラフにしなさい、、、という感じですが。。。。いい結果が出れば、見せ方考えます。
備忘(本人用)
コマンド例示
python train.py --model FCN-8s --base_model ResNet50
python train.py --model UNet --base_model ResNet152 <---NG
python train.py --model UNet --base_model VGG19 --h_flip True --rotation 15 --channel_shift 0.1 --data_aug_rate 0.5
python train.py --model DeepLabV3Plus --base_model ResNet152 --h_flip True --rotation 15 --channel_shift 0.1 --data_aug_rate 0.5
python train.py --model UNet --base_model VGG19 --h_flip True --rotation 15 --channel_shift 0.1 --data_aug_rate 0.5
Author And Source
この問題について(Semantic Segmentationのいろんなモデルが扱えるgithubがあったので、それで、性能出ししてみる。DeepLabV3+とか。), 我々は、より多くの情報をここで見つけました https://qiita.com/hitonimonarukurohyo/items/ad94d2855c390c6db2a3著者帰属:元の著者の情報は、元のURLに含まれています。著作権は原作者に属する。
Content is automatically searched and collected through network algorithms . If there is a violation . Please contact us . We will adjust (correct author information ,or delete content ) as soon as possible .