Where can I find the key and value of the feature code of the labeled first frame in your test code? about stcn HOT 7 CLOSED

1. frame_idx is just 0, representing the first frame most of the time. That interface is used to adapt to interactive/YouTubeVOS testing which is more complicated. From the propagation perspective, it is always the "first frame" in the propagation sequence.
2. The third dimension is time (B*C*T*H*W consistently in most places). It is just 1, representing a single frame.

from stcn.

1. frame_idx is just 0, representing the first frame most of the time. That interface is used to adapt to interactive/YouTubeVOS testing which is more complicated. From the propagation perspective, it is always the "first frame" in the propagation sequence.
2. The third dimension is time (B*C*T*H*W consistently in most places). It is just 1, representing a single frame.
   @hkchengrex Thanks for your reply. In the YouTube dataset the first frame is labeled, right? How do I find labeled frames in test code?

from stcn.

It is more complicated than that in YouTubeVOS. Check the dataset/our inference code for details.

from stcn.

It is more complicated than that in YouTubeVOS. Check the dataset/our inference code for details.

@hkchengrex One thing I am confused about is how to store the value of frame 0 in memory if there is no GT on frame 0?Because value holds the features and the mask.

from stcn.

@longmalongma Look at the youtube evaluation script for details. In short, we just skip the frames before the one with GT.

from stcn.

@longmalongma Look at the youtube evaluation script for details. In short, we just skip the frames before the one with GT.

@hkchengrex I seem to understand. The key_k and key_v here may not be frame 0, but they must be frames with labels?

from stcn.

@longmalongma Yes.

from stcn.