search

ubicomplab / rPPG-Toolbox

rPPG-Toolbox: Deep Remote PPG Toolbox (NeurIPS 2023)
https://arxiv.org/abs/2210.00716
Other
482 stars 117 forks source link

Pseudo ppg label function #162

Closed girishvn closed 1 year ago

girishvn commented 1 year ago

Train: PURE, Test: UBFC-rPPG, Model: TSCAN FFT MAE (FFT Label): 1.3392857142857142 +/- 0.42190125596543704 FFT RMSE (FFT Label): 3.044620560179667 +/- 3.5261431285376057 FFT MAPE (FFT Label): 1.5575960591866085 +/- 0.5049767631839935 FFT Pearson (FFT Label): 0.988079459671649 +/- 0.02434079978913088

Train: PURE Pseudo Labels, Test: UBFC-rPPG GT Labels, Model: TSCAN FFT MAE (FFT Label): 9.375 +/- 2.5283082691165326 FFT RMSE (FFT Label): 18.87773870342228 +/- 142.48195440534639 FFT MAPE (FFT Label): 13.307904748480684 +/- 4.077642607327307 FFT Pearson (FFT Label): 0.582888550823001 +/- 0.12847576982834757 FFT SNR (FFT Label): -1.8506740608244514 +/- 1.7017036801128422

yahskapar commented 1 year ago

Does the current logic and the absence of a USE_PSUEDO_PPG_LABEL for the UNSUPERVISED config portion of config.py potentially throw an error when any of the dataloaders are used for unsupervised methods? LGTM aside from that.

girishvn commented 1 year ago

Edited commit message. Merging.