debug in multiprocessing

[Hello526]

Hi,

I want to step into the coco.py(../sample) to see the kp_detection function during training, learn how to generate labels (heatmaps, regrs, tags).

https://github.com/princeton-vl/CornerNet/blob/e5c39a31a8abef5841976c8eab18da86d6ee5f9a/sample/coco.py#L49

But I seem to have encountered problems that cannot be debugged in multiple processes.

Do you know how to enter the kp_detection function to see the results of each step? I want to know how the labels(heatmaps, regrs, tags) is generated.

[Hello526]

https://github.com/princeton-vl/CornerNet/blob/e5c39a31a8abef5841976c8eab18da86d6ee5f9a/sample/coco.py#L51

I created a breakpoint in the above position, and the program stays in the position of the picture and cannot continue to run.

[Hello526]

[Hello526]

I don't know how to debug in multiple processes.

[YJLCV]

Have you tried too many GPU training? For example, training CornerNet with four 1080ti sheets

[Hanawh]

I set all xxx.start() to xxx.run() in train.py to debug.