Train models
python train.py --n_epochs 50 --Model LSTM --GAN WGAN_GP --kwidth 7 --n_fc 16 --data_path path_to_dataset --save_model_path path_to_save_model --emb_size 512 --num_layers 3 --num_heads 8 --lr 0.0001 --Dis WFAN_GP --dis_loss WGAN_GP --sample_interval 1 --Loss exp --study experiment_name
For the important paramaters
--n_epochs: The number of training epochs --Model: The training networks, including CNN_auto,LSTM,GAN,Transformer --GAN: Choose the GAN model if the --Model is GAN, including CGAN,WGAN_GP --kwidth: The number of kernel width for conv1d --n_fc: The number of start kernels for conv1d --data_path: The path to /Datasets --emb_size: The size of emb_size for Multi_attn and hidden_layer for LSTM --lr: Learning rate --Dis: The network structure for discriminator in GAN, including WGAN_GP,SN,normal --dis_loss: The loss function for discriminator, WGAN_GP or hinge --Loss: The loss function for CNN_auto, LSTM and transformer, including L1, L2, exp
Test models
python test.py --Model_file path_to_model_input --study experiment_name
Information in model_input.txt
model_name(CNN-AE/LSTM/GAN/Transformer) The path to the .pth file model_name
For example: CNN-AE /edahome/pcslab/pcs04/RPPG/rPPG/Best/CNN_auto/pos_CNN_auto_lr001_n16_b16_k7_layers3_d1/save_10_best.pth CNN-AE LSTM /edahome/pcslab/pcs04/RPPG/rPPG/Best/LSTM/L2/save_17_best.pth LSTM GAN /edahome/pcslab/pcs04/RPPG/rPPG/Best/GAN/best/pos_SN_lr0001_n16_b4_k7_layers6_d1/save_42.pth GAN Transformer /edahome/pcslab/pcs04/RPPG/rPPG/Best/pos_Transformer_k31_emb128_head4_layers6_n16_b32_exp03/save_36_best.pth Transformer