WPC和WPC2.0训练参数

你好，请参考repo内的train.sh的参数配置。

具体如下：

CUDA_VISIBLE_DEVICES=0 python -u train.py \ --learning_rate 0.00005 \ --model MM_PCQA \ --batch_size 8 \ --database SJTU \ --data_dir_2d path_to_sjtu_projections/ \ --data_dir_pc path_to_sjtu_patch_2048/ \ --loss l2rank \ --num_epochs 50 \ --k_fold_num 9 \

logs/sjtu_mmpcqa.log

CUDA_VISIBLE_DEVICES=0 python -u train.py \ --learning_rate 0.00005 \ --model MM_PCQA \ --batch_size 8 \ --database WPC \ --data_dir_2d path_to_wpc_projections/ \ --data_dir_pc path_to_wpc_patch_2048/ \ --loss l2rank \ --num_epochs 50 \ --k_fold_num 5 \

logs/wpc_mmpcqa.log

CUDA_VISIBLE_DEVICES=0 python -u train.py \ --learning_rate 0.00005 \ --model MM_PCQA \ --batch_size 8 \ --database WPC2.0 \ --data_dir_2d path_to_wpc2.0_projections/ \ --data_dir_pc path_to_wpc2.0_patch_2048/ \ --loss l2rank \ --num_epochs 50 \ --k_fold_num 4 \

logs/wpc2.0_mmpcqa.log

我这里可以提供WPC的训练log供你参考：

load the pretrained model, done！ Using model: MM-PCQA Using l2rank loss Using Adam optimizer, initial learning rate: 5e-05 Ready to train network

Epoch 1 averaged training loss: 911.0889 The current learning rate is 0.000050 Epoch 1 training time cost: 66.6278 seconds Test results: SROCC=0.7021, KROCC=0.5256, PLCC=0.7007, RMSE=16.6320 Update best model using best_val_criterion Update the best Test results: SROCC=0.7021, KROCC=0.5256, PLCC=0.7007, RMSE=16.6320 Epoch 2 averaged training loss: 291.5356 The current learning rate is 0.000050 Epoch 2 training time cost: 66.6125 seconds Test results: SROCC=0.7268, KROCC=0.5470, PLCC=0.7327, RMSE=15.8665 Update best model using best_val_criterion Update the best Test results: SROCC=0.7268, KROCC=0.5470, PLCC=0.7327, RMSE=15.8665 Epoch 3 averaged training loss: 249.2855 The current learning rate is 0.000050 Epoch 3 training time cost: 66.6923 seconds Test results: SROCC=0.7523, KROCC=0.5593, PLCC=0.6807, RMSE=17.0792 Update best model using best_val_criterion Update the best Test results: SROCC=0.7523, KROCC=0.5593, PLCC=0.6807, RMSE=17.0792 Epoch 4 averaged training loss: 228.0509 The current learning rate is 0.000050 Epoch 4 training time cost: 66.7481 seconds Test results: SROCC=0.7922, KROCC=0.6071, PLCC=0.7982, RMSE=14.0444 Update best model using best_val_criterion Update the best Test results: SROCC=0.7922, KROCC=0.6071, PLCC=0.7982, RMSE=14.0444 Epoch 5 averaged training loss: 239.8473 The current learning rate is 0.000050 Epoch 5 training time cost: 66.9706 seconds Test results: SROCC=0.7270, KROCC=0.5273, PLCC=0.7362, RMSE=15.7769 Epoch 6 averaged training loss: 233.9770 The current learning rate is 0.000050 Epoch 6 training time cost: 66.7346 seconds Test results: SROCC=0.7945, KROCC=0.6091, PLCC=0.7928, RMSE=14.2077 Update best model using best_val_criterion Update the best Test results: SROCC=0.7945, KROCC=0.6091, PLCC=0.7928, RMSE=14.2077 Epoch 7 averaged training loss: 187.2974 The current learning rate is 0.000050 Epoch 7 training time cost: 66.5150 seconds Test results: SROCC=0.8089, KROCC=0.6194, PLCC=0.8022, RMSE=13.9183 Update best model using best_val_criterion Update the best Test results: SROCC=0.8089, KROCC=0.6194, PLCC=0.8022, RMSE=13.9183 Epoch 8 averaged training loss: 170.0297 The current learning rate is 0.000045 Epoch 8 training time cost: 66.9894 seconds Test results: SROCC=0.7255, KROCC=0.5383, PLCC=0.7194, RMSE=16.1920 Epoch 9 averaged training loss: 173.7674 The current learning rate is 0.000045 Epoch 9 training time cost: 66.8328 seconds Test results: SROCC=0.8036, KROCC=0.6075, PLCC=0.8009, RMSE=13.9586 Epoch 10 averaged training loss: 184.6821 The current learning rate is 0.000045 Epoch 10 training time cost: 66.5761 seconds Test results: SROCC=0.7716, KROCC=0.5825, PLCC=0.7786, RMSE=14.6281 Epoch 11 averaged training loss: 192.9524 The current learning rate is 0.000045 Epoch 11 training time cost: 66.6049 seconds Test results: SROCC=0.8359, KROCC=0.6549, PLCC=0.8369, RMSE=12.7607 Update best model using best_val_criterion Update the best Test results: SROCC=0.8359, KROCC=0.6549, PLCC=0.8369, RMSE=12.7607 Epoch 12 averaged training loss: 192.2851 The current learning rate is 0.000045 Epoch 12 training time cost: 67.1979 seconds Test results: SROCC=0.7492, KROCC=0.5609, PLCC=0.7596, RMSE=15.1623 Epoch 13 averaged training loss: 178.3933 The current learning rate is 0.000045 Epoch 13 training time cost: 66.6491 seconds Test results: SROCC=0.8248, KROCC=0.6325, PLCC=0.8259, RMSE=13.1455 Epoch 14 averaged training loss: 192.9511 The current learning rate is 0.000045 Epoch 14 training time cost: 67.0698 seconds Test results: SROCC=0.7658, KROCC=0.5764, PLCC=0.7754, RMSE=14.7205 Epoch 15 averaged training loss: 167.6623 The current learning rate is 0.000045 Epoch 15 training time cost: 66.6250 seconds Test results: SROCC=0.7778, KROCC=0.5830, PLCC=0.7744, RMSE=14.7504 Epoch 16 averaged training loss: 160.3173 The current learning rate is 0.000041 Epoch 16 training time cost: 67.1622 seconds Test results: SROCC=0.7971, KROCC=0.5994, PLCC=0.8032, RMSE=13.8869 Epoch 17 averaged training loss: 171.5947 The current learning rate is 0.000041 Epoch 17 training time cost: 67.0610 seconds Test results: SROCC=0.8219, KROCC=0.6292, PLCC=0.8225, RMSE=13.2580 Epoch 18 averaged training loss: 173.9903 The current learning rate is 0.000041 Epoch 18 training time cost: 66.6623 seconds Test results: SROCC=0.8238, KROCC=0.6303, PLCC=0.8367, RMSE=12.7681 Epoch 19 averaged training loss: 159.9130 The current learning rate is 0.000041 Epoch 19 training time cost: 67.3795 seconds Test results: SROCC=0.7943, KROCC=0.5968, PLCC=0.7918, RMSE=14.2405 Epoch 20 averaged training loss: 171.1829 The current learning rate is 0.000041 Epoch 20 training time cost: 67.0669 seconds Test results: SROCC=0.8470, KROCC=0.6654, PLCC=0.8480, RMSE=12.3568 Update best model using best_val_criterion Update the best Test results: SROCC=0.8470, KROCC=0.6654, PLCC=0.8480, RMSE=12.3568 Epoch 21 averaged training loss: 170.1593 The current learning rate is 0.000041 Epoch 21 training time cost: 66.8068 seconds Test results: SROCC=0.8005, KROCC=0.6227, PLCC=0.8058, RMSE=13.8051 Epoch 22 averaged training loss: 161.5014 The current learning rate is 0.000041 Epoch 22 training time cost: 66.7364 seconds Test results: SROCC=0.7940, KROCC=0.5979, PLCC=0.7928, RMSE=14.2085 Epoch 23 averaged training loss: 174.9371 The current learning rate is 0.000041 Epoch 23 training time cost: 67.0592 seconds Test results: SROCC=0.7782, KROCC=0.5865, PLCC=0.7869, RMSE=14.3860 Epoch 24 averaged training loss: 137.6525 The current learning rate is 0.000036 Epoch 24 training time cost: 66.7049 seconds Test results: SROCC=0.8457, KROCC=0.6634, PLCC=0.8434, RMSE=12.5262 Epoch 25 averaged training loss: 148.3116 The current learning rate is 0.000036 Epoch 25 training time cost: 66.5353 seconds Test results: SROCC=0.7833, KROCC=0.6029, PLCC=0.7972, RMSE=14.0756 Epoch 26 averaged training loss: 190.7232 The current learning rate is 0.000036 Epoch 26 training time cost: 66.7023 seconds Test results: SROCC=0.8636, KROCC=0.6760, PLCC=0.8697, RMSE=11.5069 Update best model using best_val_criterion Update the best Test results: SROCC=0.8636, KROCC=0.6760, PLCC=0.8697, RMSE=11.5069 Epoch 27 averaged training loss: 140.3431 The current learning rate is 0.000036 Epoch 27 training time cost: 66.7309 seconds Test results: SROCC=0.8423, KROCC=0.6494, PLCC=0.8445, RMSE=12.4836 Epoch 28 averaged training loss: 137.8987 The current learning rate is 0.000036 Epoch 28 training time cost: 67.3802 seconds Test results: SROCC=0.8213, KROCC=0.6352, PLCC=0.8197, RMSE=13.3550 Epoch 29 averaged training loss: 123.6086 The current learning rate is 0.000036 Epoch 29 training time cost: 66.9521 seconds Test results: SROCC=0.8210, KROCC=0.6380, PLCC=0.8225, RMSE=13.2596 Epoch 30 averaged training loss: 132.0040 The current learning rate is 0.000036 Epoch 30 training time cost: 66.4477 seconds Test results: SROCC=0.8530, KROCC=0.6657, PLCC=0.8559, RMSE=12.0553 Epoch 31 averaged training loss: 139.0212 The current learning rate is 0.000036 Epoch 31 training time cost: 66.8905 seconds Test results: SROCC=0.8216, KROCC=0.6314, PLCC=0.8247, RMSE=13.1849 Epoch 32 averaged training loss: 156.2805 The current learning rate is 0.000033 Epoch 32 training time cost: 66.9564 seconds Test results: SROCC=0.8371, KROCC=0.6505, PLCC=0.8420, RMSE=12.5757 Epoch 33 averaged training loss: 134.1225 The current learning rate is 0.000033 Epoch 33 training time cost: 67.1574 seconds Test results: SROCC=0.7887, KROCC=0.6089, PLCC=0.7962, RMSE=14.1041 Epoch 34 averaged training loss: 129.2047 The current learning rate is 0.000033 Epoch 34 training time cost: 67.1867 seconds Test results: SROCC=0.8232, KROCC=0.6415, PLCC=0.8273, RMSE=13.0963 Epoch 35 averaged training loss: 123.3054 The current learning rate is 0.000033 Epoch 35 training time cost: 66.6079 seconds Test results: SROCC=0.8480, KROCC=0.6588, PLCC=0.8525, RMSE=12.1871 Epoch 36 averaged training loss: 128.1395 The current learning rate is 0.000033 Epoch 36 training time cost: 67.1524 seconds Test results: SROCC=0.8293, KROCC=0.6509, PLCC=0.8352, RMSE=12.8207 Epoch 37 averaged training loss: 122.1962 The current learning rate is 0.000033 Epoch 37 training time cost: 66.9688 seconds Test results: SROCC=0.8542, KROCC=0.6828, PLCC=0.8614, RMSE=11.8403 Epoch 38 averaged training loss: 112.3159 The current learning rate is 0.000033 Epoch 38 training time cost: 66.4609 seconds Test results: SROCC=0.8501, KROCC=0.6727, PLCC=0.8504, RMSE=12.2676 Epoch 39 averaged training loss: 121.3407 The current learning rate is 0.000033 Epoch 39 training time cost: 66.8107 seconds Test results: SROCC=0.8391, KROCC=0.6481, PLCC=0.8322, RMSE=12.9270 Epoch 40 averaged training loss: 121.6807 The current learning rate is 0.000030 Epoch 40 training time cost: 66.9426 seconds Test results: SROCC=0.8430, KROCC=0.6591, PLCC=0.8376, RMSE=12.7356 Epoch 41 averaged training loss: 127.3399 The current learning rate is 0.000030 Epoch 41 training time cost: 66.4336 seconds Test results: SROCC=0.7146, KROCC=0.5541, PLCC=0.7230, RMSE=16.1066 Epoch 42 averaged training loss: 129.5199 The current learning rate is 0.000030 Epoch 42 training time cost: 66.7883 seconds Test results: SROCC=0.8156, KROCC=0.6190, PLCC=0.8215, RMSE=13.2940 Epoch 43 averaged training loss: 134.2851 The current learning rate is 0.000030 Epoch 43 training time cost: 66.7571 seconds Test results: SROCC=0.8597, KROCC=0.6823, PLCC=0.8629, RMSE=11.7816 Epoch 44 averaged training loss: 120.7997 The current learning rate is 0.000030 Epoch 44 training time cost: 67.1869 seconds Test results: SROCC=0.8437, KROCC=0.6637, PLCC=0.8511, RMSE=12.2390 Epoch 45 averaged training loss: 106.8727 The current learning rate is 0.000030 Epoch 45 training time cost: 67.0713 seconds Test results: SROCC=0.8436, KROCC=0.6578, PLCC=0.8472, RMSE=12.3848 Epoch 46 averaged training loss: 108.2125 The current learning rate is 0.000030 Epoch 46 training time cost: 67.3196 seconds Test results: SROCC=0.8000, KROCC=0.6018, PLCC=0.8073, RMSE=13.7565 Epoch 47 averaged training loss: 126.2806 The current learning rate is 0.000030 Epoch 47 training time cost: 67.5054 seconds Test results: SROCC=0.8154, KROCC=0.6288, PLCC=0.8176, RMSE=13.4237 Epoch 48 averaged training loss: 115.5771 The current learning rate is 0.000027 Epoch 48 training time cost: 67.2151 seconds Test results: SROCC=0.8221, KROCC=0.6312, PLCC=0.8202, RMSE=13.3359 Epoch 49 averaged training loss: 96.0818 The current learning rate is 0.000027 Epoch 49 training time cost: 67.3727 seconds Test results: SROCC=0.8270, KROCC=0.6420, PLCC=0.8308, RMSE=12.9743 Epoch 50 averaged training loss: 90.8436 The current learning rate is 0.000027 Epoch 50 training time cost: 66.6264 seconds Test results: SROCC=0.8439, KROCC=0.6634, PLCC=0.8479, RMSE=12.3596

zzc-1998 / MM-PCQA

WPC和WPC2.0训练参数 #3