chen wei
chen wei
Ｔhe ` train_nets_mgpu.py` have some bugs, you should do some modify before using this code. About how to modify the code, you could reference the code `train_nets.py`.
The author use 100 layer network. Please compare with the `LResNet50E-IR`. Another reason is the author using a large `batchsize` which I can't use because of the memory limit.
I'm not quite sure, but I think you can find some tips from this [tutorials](https://tensorpack.readthedocs.io/tutorial/input-source.html) .
`embeddings = tf.get_default_graph().get_tensor_by_name('resnet_v1_50/E_BN2/Identity:0')`
I think it means if theta ｅｘｃeed the range of [0, pi], the `arcface` changed into `cos face`. This code is referenced from official implementation.
Please reference `readme.md ` file `Training Tips` part. The original model contains some bug. The modified model only `160M`.
The function `cosineface_losses` in `face_losses.py` is implementation of `CosFace: Large Margin Cosine Loss for Deep Face Recognition`.
In `tensorflow`, element-wise assign value to tensor is not allowed. If you want to do element-wise assign you can reference the `arcface_loss` or `cosineface_losses` which contains code for assignment by...
` images[k][i][j][0] = 0 images[k][i][j][1] = 0 images[k][i][j][2] = 0 ` This is not allowed in `tensorflow`. 1. You should build a mask 2. Set the element in this mask...
Can you paste your results?