Rex Cheng

I don't think you can simply do that. STCN needs multiple forward passes and those depend on a growing memory bank. One possible way is to compute the FLOPs for...

I like the output of 0.4 more though, guess I will compute the centroid again in numpy for now.

Stay safe! I am in Hong Kong so 👀

Thanks a lot. Is it possible to 1. Remove formatting changes like ' -> "? It would make the commits cleaner. 2. Move "bulk interactive" as an argument to the...

I am not using any formatter. Just trying to reduce unnecessary diff. Thanks.

@mjw123bs If you are not going to use them then no.

@mjw123bs 1. I don't know how is your data formatted. I imagine you can copy/paste/edit some of our loading code in `train.py` and `dataset/vos_dataset.py`. 2. Yes. Like DAVIS/YouTubeVOS.

I couldn't recall but I used fusion_data. I don't have that on my computer right now but if that's what you saw it should be correct.

If I zoom into the image I see a lot of noises and artifacts. No surprise to me that it kind of failed in this case.

It was a bug. I fixed it. Can you pull and see if it works?