Deepinx

The training and test datasets are obtained from insightface repo. During training, data augmentation is used with a random combination of horizontal flip, rotation (+/- 40 degrees), and scaling (0.8...

This problem has been fixed on recent updates. Thank you a lot for your feedback. Best Regards.

I just set `network.sdu.net_coherent = True` and revise line 579 of sym_heatmap.py to `coherent_weight = 0.001`, it seems this nan problem can be solved.

I did this following the guides of the origial paper, as in the paper `Therefore, we employ the CE loss for Lp-g and the MSE loss for Lp-p, respectively. l...

What batch size and lr do you use? You can try different batch size or lr, perhaps it can solve your problem.

I used batch-size 16 and lr 0.00002 at the first several epochs. The spike did not appear. You can try the following commands: ``` NETWORK='sdu' MODELDIR='./model_2d' mkdir -p "$MODELDIR" PREFIX="$MODELDIR/$NETWORK"...

"Inside Transformer" and "Outside Transformer" are still being tested and we have found that these transformers cannot improve performance as stated in the original paper. We will provide the code...

What do you mean by multi class classification? This method is designed for face detection and alignment not for classification.

In this method, age and gender are output at the same layer and trained at the same time, so for the time being I have not figured out a way...

Facial landmarks have been added to the original SSH method, so landmarks can be directly derived from it. The accuracy was copied from mxnet-ssh. I did not verify it myself....