Laughing

> This issue is not present in coco dataset, it is happening when fine-tuning on a custom dataset. @bit-scientist @fcakyon hi, can you guys provide me more info about your...

@bit-scientist hi I've reproduced your results, which I made a PR #490 to fix it. It'll be merged later today.

@AyushExel @glenn-jocher Hey I think probably it's better to put `Result` in `self.postprocess` like this:  for four reasons: 1. there're some hardcode about how we split outputs in our...

@AyushExel oh I haven't started yet. it's perfect if you'd like to start. ^ ^

@AyushExel oh the batch-size can exceed 1 if the source is webcam. But actually I think we don't need idx in `result` anyway, we just need idx in function `write_result`....

@AyushExel perhaps we can do something like this in postprocess: ```python return [Result(...) for pred in preds] ``` we'll still return a list like before, instead we have Result as...

@AyushExel @glenn-jocher hey now basically `Result` and `np/pil` are working correctly as we expected. TODO: - more tests - review and clean the code - update Docs - probably update...

@glenn-jocher @AyushExel ok I'm trying to merge `img` and `source`, `inference` and `__call__`. The latter could take a while.

@glenn-jocher @AyushExel ok I've unified the `source`,`img` and `inference`,`__call__`. Also updated Docs:   probably we should independent the `Result usage` part outside this block... as it's possible that users...

@universea @Wanghe1997 是的你计算了所有的时间，包括读取图片，前处理推理后处理，绘制的时间。 @Wanghe1997 log中的postprocess就是nms的时间 @universea 是的目前还不支持numpy作为输入，但这个在我们TODO列表里，我们会尽快支持的。:)