zlai0

Hi Shu, Sure, thanks for your quick reply. I will try that. But I was just wondering how do you dilate the joint, and how is the voting done?

Hi Shu, Cool. Thanks for your explanation!

I would say the easiest thing to try first is to use CPU for inference. Another possibility is to reduce the number of channels (i.e. the maximum number of object...

You can visualise the output similar to how you visualise image segmentation (COCO, etc.) results. I believe I used cv2.drawContours and also some other functions. Something like this might also...

@houhouhouhou11 "Only short" should be 1 frame. During testing process each label is propagated separately (just like each of three color channels).

> I have same question, because when I evaluate with 1 reference, i have bad result. I demo 2 sequence frame below Could you first try evaluate J&F to see...

The module computes similarities between a centre pixel and its local/neighbour pixels.

You can add an extra option `--ref 2` when you test.

This is to enable multi-object tracking in a single inference's time (DAVIS, Youtube-VOS). Basically each one-hot label represents an instance label.

Sorry I still have no idea.. We had another independent reimplementation which also only runs on pytorch