search

ubicomplab / rPPG-Toolbox

rPPG-Toolbox: Deep Remote PPG Toolbox (NeurIPS 2023)
https://arxiv.org/abs/2210.00716
Other
459 stars 112 forks source link

ERROR: No Face Detected #268

Closed 408550969 closed 5 months ago

408550969 commented 5 months ago

Hi, when I run the test set and report Error: No Face Detected twice, will these two folders that did not detect faces be taken for testing, and will this affect the MAE?

yahskapar commented 5 months ago

Yes, they will still be included in the test split. Due to the face detection failure, the only difference is that the whole frame of the original video will be used and resized to whatever resolution specified (e.g., 72x72) rather than just the face region.

Out of curiosity, what dataset are you encountering this error with? In your config, are you using the default Haar Cascade (HC) approach for face detection? If so, maybe try RetinaFace (RF) instead. RetinaFace should generally allow for better face detection results, but part of the reason it isn't used by default is the benchmark results in the NeurIPS 2023 paper all happened to use Haar Cascade.

408550969 commented 5 months ago

Thank you, I roughly understand why the MAE has reached such a high of 7. I used some infrared videos I collected myself, and since the camera is tilted, the face I obtained has a certain angle. I used default face detection