Feng Wang comments

Results 182 comments of


                                            Feng Wang

KeyError: 'model'

`-c weights/yolox_s.pth` is not needed here, and `yolox_s.pth` not a checkpoint file, it only contains model state.

请问yolox_tiny的cfg文件可以从哪里获得呢

I'm not so familar with nnpack. In my opinion, such cfg file should not be a python file?

多GPU训练验证完卡死的情况

Is there any eval result logged out? Which epoch during training are your process hangs ?

多GPU训练验证完卡死的情况

What if you try solution in #1316 @YuNaruto @yannixzhu

多GPU训练验证完卡死的情况

You might try `pip install -v -e .` first. @yannixzhu

create EXP file for coco dataset

You might refer code [here](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/exp/yolox_base.py#L41-L46) for your own coco format dataset. If you like, override `get_dataloader` method in `Exp` class could also help you.

draw debug image

Sorry, I couldn't catch the meaning of `draw debug image`. Are you mean `visualize image after transform` or `visualize prediction of image` or `visual prediction and ground truth of image`?

multi GPU training gets stuck after first validation

I think stucking during traning might be caused by long time compile and evaluation on master process only. Is it producible and what's the platform are you using? WSL?

multi GPU training gets stuck after first validation

duplicated with #1303

multi GPU training gets stuck after first validation

@paroj One possible way to workaround this issue is to compile the op at the begining (`python3 -m pip install -v -e .` first). It will compile fast_cocoeval and loading...