search

Tencent / TFace

A trusty face analysis research platform developed by Tencent Youtu Lab
Apache License 2.0
1.31k stars 226 forks source link

Privacy-Preserving Face Recognition Implementation #98

Open Vinayak-Shrivastava opened 1 year ago

Vinayak-Shrivastava commented 1 year ago

I am currently undertaking the implementation of the paper titled "Privacy-Preserving Face Recognition with Learnable Privacy Budgets in Frequency Domain." Before anything else, I would like to extend my heartfelt congratulations to the remarkable researchers involved for their pioneering work and the subsequent publication. My goal is to generate hashed formats of face images, which can be employed for facial recognition in various practical applications. However, I am encountering some challenges with the code implementation and would greatly appreciate guidance and assistance on how to proceed.

I have chosen to carry out this implementation within a Google Colab notebook. To work with the task at hand, I have uploaded custom images and adjusted the batch size accordingly.

Since Colab provides a single GPU, I have been training the model using only one GPU, with the num_workers parameter set to 1 in the YAML file. Unfortunately, while running the training script, I encountered the following error in the log file:

RuntimeError: DataLoader worker (pid 57549) is killed by signal: Aborted. 
RuntimeError: DataLoader worker (pid(s) 57549) exited unexpectedly
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 57495) of binary: /usr/bin/python3

Considering the challenge of implementing a multi-processing script with a single GPU, I thought that setting num_workers to 0 might serve as a potential workaround. However, this adjustment resulted in the following error:

ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: -6) local_rank: 0 (pid: 39543) of binary: /usr/bin/python3

Please advice on how I can resolve this.

Furthermore, I am immensely intrigued by the process described in the paper for generating hashed representations of facial images. The potential applications stemming from this technique appear boundless, intensifying my curiosity and eagerness to explore this area further.

Thanks in advance.

StrugglingForBetter commented 1 year ago

Hi, I met the same problem. Did you find the solution?

Vinayak-Shrivastava commented 1 year ago

@StrugglingForBetter Not yet!!

Vinayak-Shrivastava commented 1 year ago

@StrugglingForBetter Since we are both facing the same issue, I was wondering if we could connect in some way so that we can discuss it and help each other. If you feel comfortable please reach me via mail at shrivastava.vinayak.iitkgp@gmail.com