search

thodorii / Thesis

0 stars 0 forks source link

Polished-Night : UNet Lee-Filter #4

Open thodorii opened 1 year ago

thodorii commented 1 year ago

Polished-Night

Train models after fixing the lee-filter to be channelwise.

Hyperparameters specifications:

Filters = 64
Epochs = 30 
LR = 0.0001 (1e-4)
Scheduler = tf.keras.optimizers.schedules.ExponentialDecay(
            FLAGS.lr,
            decay_steps=200,
            decay_rate=0.96,
            staircase=True
        )

Optimizer = tf.keras.optimizers.Adam(
            learning_rate=lr_schedule,
            beta_1=0.9,
            beta_2=0.999,
            epsilon=1e-07,
            amsgrad=False,
            name='Adam',
        )
thodorii commented 1 year ago

Polished-Night S1

Water IoU: 58.564 Water Precision: 70.305 Water Recall: 77.812 Water F1: 73.868

592/592 [==============================] - 318s 498ms/step - loss: 36.8753 - mean_io_u: 0.3759 - val_loss: 11.1362 - val_mean_io_u: 0.4728
Epoch 2/30
592/592 [==============================] - 291s 492ms/step - loss: 7.8144 - mean_io_u: 0.4361 - val_loss: 5.7885 - val_mean_io_u: 0.5134
Epoch 3/30
592/592 [==============================] - 292s 492ms/step - loss: 4.9745 - mean_io_u: 0.4937 - val_loss: 4.2599 - val_mean_io_u: 0.5614
Epoch 4/30
592/592 [==============================] - 290s 490ms/step - loss: 3.8483 - mean_io_u: 0.5883 - val_loss: 3.3683 - val_mean_io_u: 0.7086
Epoch 5/30
592/592 [==============================] - 290s 490ms/step - loss: 3.1561 - mean_io_u: 0.7092 - val_loss: 2.9045 - val_mean_io_u: 0.7051
Epoch 6/30
592/592 [==============================] - 289s 487ms/step - loss: 2.7655 - mean_io_u: 0.7246 - val_loss: 2.5800 - val_mean_io_u: 0.7387
Epoch 7/30
592/592 [==============================] - 289s 487ms/step - loss: 2.4898 - mean_io_u: 0.7279 - val_loss: 2.3542 - val_mean_io_u: 0.6977
Epoch 8/30
592/592 [==============================] - 288s 486ms/step - loss: 2.2777 - mean_io_u: 0.7300 - val_loss: 2.1646 - val_mean_io_u: 0.7059
Epoch 9/30
592/592 [==============================] - 288s 485ms/step - loss: 2.1102 - mean_io_u: 0.7311 - val_loss: 2.0171 - val_mean_io_u: 0.7057
Epoch 10/30
592/592 [==============================] - 289s 487ms/step - loss: 1.9757 - mean_io_u: 0.7319 - val_loss: 1.8972 - val_mean_io_u: 0.7056
Epoch 11/30
592/592 [==============================] - 289s 486ms/step - loss: 1.8662 - mean_io_u: 0.7327 - val_loss: 1.8004 - val_mean_io_u: 0.7028
Epoch 12/30
592/592 [==============================] - 288s 485ms/step - loss: 1.7762 - mean_io_u: 0.7334 - val_loss: 1.7216 - val_mean_io_u: 0.6978
Epoch 13/30
592/592 [==============================] - 288s 486ms/step - loss: 1.7015 - mean_io_u: 0.7343 - val_loss: 1.6571 - val_mean_io_u: 0.6916
Epoch 14/30
592/592 [==============================] - 288s 486ms/step - loss: 1.6394 - mean_io_u: 0.7351 - val_loss: 1.6029 - val_mean_io_u: 0.6873
Epoch 15/30
592/592 [==============================] - 288s 486ms/step - loss: 1.5868 - mean_io_u: 0.7360 - val_loss: 1.5557 - val_mean_io_u: 0.6865
Epoch 16/30
592/592 [==============================] - 288s 485ms/step - loss: 1.5424 - mean_io_u: 0.7367 - val_loss: 1.5143 - val_mean_io_u: 0.6884
Epoch 17/30
592/592 [==============================] - 288s 485ms/step - loss: 1.5047 - mean_io_u: 0.7373 - val_loss: 1.4782 - val_mean_io_u: 0.6919
Epoch 18/30
592/592 [==============================] - 288s 485ms/step - loss: 1.4726 - mean_io_u: 0.7378 - val_loss: 1.4468 - val_mean_io_u: 0.6962
Epoch 19/30
592/592 [==============================] - 288s 486ms/step - loss: 1.4452 - mean_io_u: 0.7382 - val_loss: 1.4198 - val_mean_io_u: 0.7006
Epoch 20/30
592/592 [==============================] - 288s 486ms/step - loss: 1.4218 - mean_io_u: 0.7387 - val_loss: 1.3967 - val_mean_io_u: 0.7047
Epoch 21/30
592/592 [==============================] - 288s 485ms/step - loss: 1.4017 - mean_io_u: 0.7390 - val_loss: 1.3769 - val_mean_io_u: 0.7085
Epoch 22/30
592/592 [==============================] - 288s 485ms/step - loss: 1.3846 - mean_io_u: 0.7394 - val_loss: 1.3601 - val_mean_io_u: 0.7115
Epoch 23/30
592/592 [==============================] - 289s 487ms/step - loss: 1.3700 - mean_io_u: 0.7397 - val_loss: 1.3457 - val_mean_io_u: 0.7145
Epoch 24/30
592/592 [==============================] - 288s 486ms/step - loss: 1.3575 - mean_io_u: 0.7399 - val_loss: 1.3333 - val_mean_io_u: 0.7173
Epoch 25/30
592/592 [==============================] - 288s 485ms/step - loss: 1.3467 - mean_io_u: 0.7402 - val_loss: 1.3229 - val_mean_io_u: 0.7194
Epoch 26/30
592/592 [==============================] - 288s 487ms/step - loss: 1.3375 - mean_io_u: 0.7404 - val_loss: 1.3138 - val_mean_io_u: 0.7219
Epoch 27/30
592/592 [==============================] - 288s 486ms/step - loss: 1.3295 - mean_io_u: 0.7406 - val_loss: 1.3060 - val_mean_io_u: 0.7239
Epoch 28/30
592/592 [==============================] - 288s 486ms/step - loss: 1.3226 - mean_io_u: 0.7408 - val_loss: 1.2993 - val_mean_io_u: 0.7255
Epoch 29/30
592/592 [==============================] - 287s 485ms/step - loss: 1.3166 - mean_io_u: 0.7409 - val_loss: 1.2936 - val_mean_io_u: 0.7267
Epoch 30/30
592/592 [==============================] - 288s 486ms/step - loss: 1.3114 - mean_io_u: 0.7411 - val_loss: 1.2887 - val_mean_io_u: 0.7276
thodorii commented 1 year ago

Polished-Night S2

Water IoU: 58.784 Water Precision: 69.373 Water Recall: 79.387 Water F1: 74.043

592/592 [==============================] - 325s 516ms/step - loss: 34.5084 - mean_io_u: 0.5028 - val_loss: 9.6149 - val_mean_io_u: 0.5761
Epoch 2/30
592/592 [==============================] - 309s 520ms/step - loss: 6.5898 - mean_io_u: 0.5755 - val_loss: 4.9874 - val_mean_io_u: 0.6054
Epoch 3/30
592/592 [==============================] - 302s 508ms/step - loss: 4.1689 - mean_io_u: 0.6437 - val_loss: 3.5733 - val_mean_io_u: 0.7762
Epoch 4/30
592/592 [==============================] - 300s 507ms/step - loss: 3.2514 - mean_io_u: 0.7198 - val_loss: 2.9451 - val_mean_io_u: 0.7837
Epoch 5/30
592/592 [==============================] - 302s 508ms/step - loss: 2.7484 - mean_io_u: 0.7270 - val_loss: 2.5294 - val_mean_io_u: 0.7830
Epoch 6/30
592/592 [==============================] - 308s 520ms/step - loss: 2.3937 - mean_io_u: 0.7311 - val_loss: 2.2248 - val_mean_io_u: 0.7805
Epoch 7/30
592/592 [==============================] - 304s 512ms/step - loss: 2.1280 - mean_io_u: 0.7337 - val_loss: 1.9937 - val_mean_io_u: 0.7785
Epoch 8/30
592/592 [==============================] - 309s 520ms/step - loss: 1.9240 - mean_io_u: 0.7353 - val_loss: 1.8151 - val_mean_io_u: 0.7778
Epoch 9/30
592/592 [==============================] - 304s 512ms/step - loss: 1.7647 - mean_io_u: 0.7365 - val_loss: 1.6750 - val_mean_io_u: 0.7790
Epoch 10/30
592/592 [==============================] - 305s 514ms/step - loss: 1.6387 - mean_io_u: 0.7376 - val_loss: 1.5635 - val_mean_io_u: 0.7809
Epoch 11/30
592/592 [==============================] - 304s 513ms/step - loss: 1.5378 - mean_io_u: 0.7387 - val_loss: 1.4739 - val_mean_io_u: 0.7825
Epoch 12/30
592/592 [==============================] - 309s 520ms/step - loss: 1.4561 - mean_io_u: 0.7395 - val_loss: 1.4011 - val_mean_io_u: 0.7836
Epoch 13/30
592/592 [==============================] - 306s 515ms/step - loss: 1.3894 - mean_io_u: 0.7402 - val_loss: 1.3413 - val_mean_io_u: 0.7842
Epoch 14/30
592/592 [==============================] - 305s 515ms/step - loss: 1.3346 - mean_io_u: 0.7408 - val_loss: 1.2919 - val_mean_io_u: 0.7842
Epoch 15/30
592/592 [==============================] - 304s 514ms/step - loss: 1.2892 - mean_io_u: 0.7414 - val_loss: 1.2508 - val_mean_io_u: 0.7835
Epoch 16/30
592/592 [==============================] - 309s 520ms/step - loss: 1.2515 - mean_io_u: 0.7420 - val_loss: 1.2164 - val_mean_io_u: 0.7824
Epoch 17/30
592/592 [==============================] - 305s 515ms/step - loss: 1.2200 - mean_io_u: 0.7424 - val_loss: 1.1876 - val_mean_io_u: 0.7809
Epoch 18/30
592/592 [==============================] - 310s 522ms/step - loss: 1.1937 - mean_io_u: 0.7429 - val_loss: 1.1636 - val_mean_io_u: 0.7796
Epoch 19/30
592/592 [==============================] - 308s 519ms/step - loss: 1.1716 - mean_io_u: 0.7433 - val_loss: 1.1435 - val_mean_io_u: 0.7784
Epoch 20/30
592/592 [==============================] - 308s 521ms/step - loss: 1.1530 - mean_io_u: 0.7436 - val_loss: 1.1266 - val_mean_io_u: 0.7776
Epoch 21/30
592/592 [==============================] - 306s 513ms/step - loss: 1.1373 - mean_io_u: 0.7439 - val_loss: 1.1123 - val_mean_io_u: 0.7772
Epoch 22/30
592/592 [==============================] - 308s 519ms/step - loss: 1.1239 - mean_io_u: 0.7441 - val_loss: 1.1002 - val_mean_io_u: 0.7769
Epoch 23/30
592/592 [==============================] - 305s 512ms/step - loss: 1.1126 - mean_io_u: 0.7444 - val_loss: 1.0899 - val_mean_io_u: 0.7767
Epoch 24/30
592/592 [==============================] - 302s 509ms/step - loss: 1.1029 - mean_io_u: 0.7445 - val_loss: 1.0812 - val_mean_io_u: 0.7764
Epoch 25/30
592/592 [==============================] - 311s 524ms/step - loss: 1.0946 - mean_io_u: 0.7447 - val_loss: 1.0736 - val_mean_io_u: 0.7763
Epoch 26/30
592/592 [==============================] - 307s 517ms/step - loss: 1.0875 - mean_io_u: 0.7449 - val_loss: 1.0672 - val_mean_io_u: 0.7760
Epoch 27/30
592/592 [==============================] - 311s 524ms/step - loss: 1.0814 - mean_io_u: 0.7450 - val_loss: 1.0616 - val_mean_io_u: 0.7758
Epoch 28/30
592/592 [==============================] - 311s 525ms/step - loss: 1.0760 - mean_io_u: 0.7451 - val_loss: 1.0567 - val_mean_io_u: 0.7757
Epoch 29/30
592/592 [==============================] - 309s 520ms/step - loss: 1.0714 - mean_io_u: 0.7453 - val_loss: 1.0525 - val_mean_io_u: 0.7756
Epoch 30/30
592/592 [==============================] - 315s 532ms/step - loss: 1.0673 - mean_io_u: 0.7454 - val_loss: 1.0489 - val_mean_io_u: 0.7756
thodorii commented 1 year ago

Polished-Night S3 Water IoU: 60.237 Water Precision: 72.246 Water Recall: 78.372 Water F1: 75.185

592/592 [==============================] - 344s 549ms/step - loss: 35.5556 - mean_io_u: 0.5021 - val_loss: 10.3314 - val_mean_io_u: 0.5943
Epoch 2/30
592/592 [==============================] - 333s 561ms/step - loss: 7.0593 - mean_io_u: 0.5787 - val_loss: 5.1637 - val_mean_io_u: 0.6531
Epoch 3/30
592/592 [==============================] - 326s 549ms/step - loss: 4.2980 - mean_io_u: 0.6403 - val_loss: 3.6137 - val_mean_io_u: 0.6544
Epoch 4/30
592/592 [==============================] - 328s 553ms/step - loss: 3.2001 - mean_io_u: 0.7254 - val_loss: 2.8398 - val_mean_io_u: 0.7406
Epoch 5/30
592/592 [==============================] - 326s 550ms/step - loss: 2.6130 - mean_io_u: 0.7373 - val_loss: 2.3785 - val_mean_io_u: 0.7505
Epoch 6/30
592/592 [==============================] - 331s 558ms/step - loss: 2.2264 - mean_io_u: 0.7386 - val_loss: 2.0576 - val_mean_io_u: 0.7568
Epoch 7/30
592/592 [==============================] - 325s 548ms/step - loss: 1.9483 - mean_io_u: 0.7397 - val_loss: 1.8217 - val_mean_io_u: 0.7600
Epoch 8/30
592/592 [==============================] - 328s 553ms/step - loss: 1.7407 - mean_io_u: 0.7406 - val_loss: 1.6423 - val_mean_io_u: 0.7602
Epoch 9/30
592/592 [==============================] - 335s 564ms/step - loss: 1.5826 - mean_io_u: 0.7417 - val_loss: 1.5036 - val_mean_io_u: 0.7582
Epoch 10/30
592/592 [==============================] - 328s 553ms/step - loss: 1.4609 - mean_io_u: 0.7428 - val_loss: 1.3968 - val_mean_io_u: 0.7558
Epoch 11/30
592/592 [==============================] - 326s 550ms/step - loss: 1.3665 - mean_io_u: 0.7439 - val_loss: 1.3139 - val_mean_io_u: 0.7538
Epoch 12/30
592/592 [==============================] - 331s 558ms/step - loss: 1.2926 - mean_io_u: 0.7447 - val_loss: 1.2491 - val_mean_io_u: 0.7521
Epoch 13/30
592/592 [==============================] - 332s 561ms/step - loss: 1.2346 - mean_io_u: 0.7454 - val_loss: 1.1980 - val_mean_io_u: 0.7509
Epoch 14/30
592/592 [==============================] - 331s 559ms/step - loss: 1.1884 - mean_io_u: 0.7459 - val_loss: 1.1574 - val_mean_io_u: 0.7500
Epoch 15/30
592/592 [==============================] - 330s 557ms/step - loss: 1.1514 - mean_io_u: 0.7464 - val_loss: 1.1247 - val_mean_io_u: 0.7492
Epoch 16/30
592/592 [==============================] - 326s 550ms/step - loss: 1.1215 - mean_io_u: 0.7469 - val_loss: 1.0982 - val_mean_io_u: 0.7482
Epoch 17/30
592/592 [==============================] - 330s 557ms/step - loss: 1.0970 - mean_io_u: 0.7473 - val_loss: 1.0765 - val_mean_io_u: 0.7475
Epoch 18/30
592/592 [==============================] - 324s 547ms/step - loss: 1.0769 - mean_io_u: 0.7477 - val_loss: 1.0586 - val_mean_io_u: 0.7477
Epoch 19/30
592/592 [==============================] - 322s 544ms/step - loss: 1.0602 - mean_io_u: 0.7480 - val_loss: 1.0436 - val_mean_io_u: 0.7488
Epoch 20/30
592/592 [==============================] - 324s 546ms/step - loss: 1.0464 - mean_io_u: 0.7483 - val_loss: 1.0310 - val_mean_io_u: 0.7506
Epoch 21/30
592/592 [==============================] - 321s 541ms/step - loss: 1.0349 - mean_io_u: 0.7486 - val_loss: 1.0205 - val_mean_io_u: 0.7529
Epoch 22/30
592/592 [==============================] - 320s 540ms/step - loss: 1.0252 - mean_io_u: 0.7488 - val_loss: 1.0119 - val_mean_io_u: 0.7548
Epoch 23/30
592/592 [==============================] - 321s 541ms/step - loss: 1.0170 - mean_io_u: 0.7490 - val_loss: 1.0046 - val_mean_io_u: 0.7563
Epoch 24/30
592/592 [==============================] - 322s 543ms/step - loss: 1.0101 - mean_io_u: 0.7492 - val_loss: 0.9985 - val_mean_io_u: 0.7574
Epoch 25/30
592/592 [==============================] - 319s 538ms/step - loss: 1.0041 - mean_io_u: 0.7494 - val_loss: 0.9933 - val_mean_io_u: 0.7581
Epoch 26/30
592/592 [==============================] - 322s 543ms/step - loss: 0.9991 - mean_io_u: 0.7496 - val_loss: 0.9889 - val_mean_io_u: 0.7587
Epoch 27/30
592/592 [==============================] - 324s 546ms/step - loss: 0.9947 - mean_io_u: 0.7498 - val_loss: 0.9850 - val_mean_io_u: 0.7591
Epoch 28/30
592/592 [==============================] - 320s 540ms/step - loss: 0.9909 - mean_io_u: 0.7499 - val_loss: 0.9817 - val_mean_io_u: 0.7595
Epoch 29/30
592/592 [==============================] - 322s 544ms/step - loss: 0.9876 - mean_io_u: 0.7500 - val_loss: 0.9788 - val_mean_io_u: 0.7597
Epoch 30/30
592/592 [==============================] - 324s 547ms/step - loss: 0.9848 - mean_io_u: 0.7501 - val_loss: 0.9763 - val_mean_io_u: 0.7597