FaceX-Zoo issues

Results 120 FaceX-Zoo issues

Sort by recently updated

Distributed FC training

To train webface 20M subjects https://www.face-benchmark.org/, we should use distributed fc training. Do you have schedule to implement it https://github.com/Tencent/TFace/tree/master/tasks/distfc

John1231983

face feature with cpu 使用torch cpu版问题

你好，我在使用的时候，torch版本用的cpu版本，运行faceSDK中 face_feature.py 在torch.load()中也指定了cpu，但是一直报错 ERROR 2021-10-12 19:47:02 face_feature.py: 59] Failed to extract facial features! ERROR 2021-10-12 19:47:02 face_feature.py: 60] module must have its parameters and buffers on device cuda:0 (device_ids[0]) but...

Lcfling

the define of kl_loss can't found when the loss_type is SKD in face_lightning.loss_def.py

Hi, the define of kl_loss can't found when the loss_type is SKD in face_lightning.loss_def.py and when i train the recognition task by mv_softmax and kl loss, the kl_loss getting lower...

niliusha123

face lightning student model

face lightning 中的学生模型无法下载，有计划开源出来吗？

YjDai

AgeDB数据集如何对齐

AgeDB数据集又68个关键点，通过crop_agedb_by_arcface.py没法跑通，原始代码里没有关于68个关键点与5个关键点之间的转换

GYC1996

AdeDB cropper not working

The AgeDB cropper provided in the [code](https://github.com/JDAI-CV/FaceX-Zoo/blob/main/test_protocol/lfw/face_cropper/crop_agedb_by_arcface.py) does not work because of the landmark size error. The cropper and dependent utilites in this library work only for 5-point, 25-point, and...

roomo7time

在SST中，为什么每个epoch训练中只使用16384个ID，而不是全量72778 IDs 的shallow data

大佬，你好。在SST中，为什么每个epoch训练中只使用16384个ID，而不是全量72778 IDs 的shallow data？这里是采用了partialfc的思想吗，还是因为这样分批处理会有其他有益的帮助？如果这里每个epoch使用全量72778 IDs 的shallow data训练，每个epoch重新shuffle一下IDs顺序，可以达到论文的精度吗？

wuyu112

Missing files for AgeDB dataset

Hi authors, Thanks so much for your work, which helps me a lot for constructing a complete custom face recognition training pipeline. I want to ask about the evaluation on...

TMaysGGS