junjiejiangjjj

Results 51 comments of junjiejiangjjj

The op https://towhee.io/video-decode/ffmpeg can load video from HTTP addresses, and image decode also support HTTP addresses.

> 个问题,如果输入是 HTTP addresses ,如何设置超时参数呀 The decoders don't support it, you can open a new issue to trace it.

We only need to support the DC which starts with `dummy_input` ends with `as_function`

if the problem occurs again, please use `stack` or `pstrace` to find where the process hangs. I have run the pipeline 1000000 times, but the problem has not occurred.

Any other infos? the stack is very similar to this https://stackoverflow.com/questions/68125101/python-multiprocessing-getting-stuck maybe because of multiprocessing.

It seems The CUDA runtime does not support the fork start method. @angelapytao https://pytorch.org/docs/stable/notes/multiprocessing.html#cuda-in-multiprocessing

![image](https://user-images.githubusercontent.com/14136703/182060689-2e248878-4964-49e7-83c5-c4aa59e08d42.png) One solution is to import torch in subprocess.

The process is blocked by queue, I change some code and the process can exit. It is just an easy fix to show you why the process is blocked. ![image](https://user-images.githubusercontent.com/14136703/182078779-5d1c9cc2-435d-41a3-a53b-9a43638448b6.png)...