Lei
Lei
Hi lyakaap, thank you for your hard work! In your code v2clean.py, **LandmarkNet** is initialized with "**_ep4_augmentation-soft_epochs-5_loss-arcface.pth_**", and then trained on GLD_v2_clean dataset. To reproduce your result, "**_ep4_augmentation-soft_epochs-5_loss-arcface.pth_**" is needed,...
The region array calculated by rmac_regions seems to be [x, y, w, h], but actually, it is used as [x, y, h, w] in RoiPooling...
There are 9958 images in datasetImages_originalSize.zip, and 8458+1000=9458 scores in AADBinfo.mat. After comparing the image names, I find that for some images, the corresponding scores cannot be found in AADBinfo.mat....