Mask visualization problem

[lzh00]

I tried to visualize the mask image with the code below:

output = 1-(occ_mask - occ_mask.min()) / (occ_mask.max() - occ_mask.min())
io.imsave(os.path.join(seq_output_folder, fname), output)

The result is not the same with your paper claimed.
Are there any problems here?

[maximelianos]

The code is misleading - MaskFlownet model does not return occlusions. If you look at the paper, this model is composed of several MaskFlownet_S, with no occlusion output in the end (figure 2 in the paper).

Thus, occlusions can be obtained only from running MaskFlownet_S, and the results are similar to the visualization.
You don't need to use normalisation there, checking min and max shows that occlusions are in range [0, 1]. Meanwhile when I ran MaskFlownet, the output was obviously optical flow, not occlusions.

[lzh00]

@lizzardhub Thank you very much. I have tried to use MaskFlownet_S configuration and model, the output occ_mask is the same with paper claimed.

[Yugt4732]

Thanks a lot. But, how can I infer in my own data by MaskFlownet_S?

@lizzardhub Thank you very much. I have tried to use MaskFlownet_S configuration and model, the output occ_mask is the same with paper claimed.

The code is misleading - MaskFlownet model does not return occlusions. If you look at the paper, this model is composed of several MaskFlownet_S, with no occlusion output in the end (figure 2 in the paper).

Thus, occlusions can be obtained only from running MaskFlownet_S, and the results are similar to the visualization.
You don't need to use normalisation there, checking min and max shows that occlusions are in range [0, 1]. Meanwhile when I ran MaskFlownet, the output was obviously optical flow, not occlusions.