ntubiolin

Hello! I really appreciate you sharing your codes and the pre-trained models! However, when using the IR-SE50 pre-trained weights to do evaluation on the Youtube Faces Database, I get only...

ERROR: packages/gizmo-agent is not a valid editable requirement

11

When I follow the readme to run the command `pip install -r requirements.txt` under `backend` directory, an error occurs: ``` ERROR: packages/gizmo-agent is not a valid editable requirement. It should...

I ran into this error: ``` File "/home/r07944011/xcos/src/model/face_recog.py", line 398, in forward index = index.type(torch.BoolTensor) AttributeError: module 'torch' has no attribute 'BoolTensor' ``` and it seems due to pytorch version...

1. Why can't I have split val loaders with independent val loaders? What if I want to not only watch val accuracies but val loss... https://github.com/ntubiolin/xcos/blob/49d3fcf5ada560106bf9083a51dffc1f3abf4d0c/src/pipeline/base_pipeline.py#L137-L142 2. Metric cannot tell...

Does BarchNorm need different weight decay?

2

In original Insight Pytorch ``` self.optimizer_fr = optim.SGD([ {'params': paras_wo_bn + [self.head.kernel], 'weight_decay': 5e-4}, {'params': paras_only_bn} ], lr = conf.lr, momentum = conf.momentum) ``` , the optimizer will apply weight_decay...