Anilkumar

@gs-ren you can find the info here https://github.com/facebookresearch/ContactPose/issues/16#issuecomment-1031823614

Hi Korrawe, Thanks for the quick response and the note. I will look forward to the models. Best, Anil

I checked the calibration folder, but there are 10 extrinsic.yml files. It is supposed to be 8, right? because we only have 8 fixed cameras. Additionally, each extrinsic.yml file has...

Thanks, @ychao-nvidia for the response. I looked in the visualization code depth = cv2.imread(depth_file, cv2.IMREAD_ANYDEPTH) --> gives depth values in 0-65535 range (uint16) and, depth = depth.astype(np.float32) / 1000. -->...

@RainerKuemmerle I have difficulty in computing the information matrix for the following example(I have vertex and edge with constraints, but I don't know how is information matrix is obtained). Could...

Thanks, @SeanChenxy for creating this issue. I am also looking for this information. @korrawe Thanks in Advance!

@SeanChenxy Did you manage to train?

"_MLP applied to each node independently_", do you mean that if we have graph of 20 nodes with labels on each node then consider these 20 nodes as different training...

yes, I just implemented. For a graph of 20 nodes(4 feature on each node), I trained two layer MLP(hidden dim=128, relu non-linearity) with MSE loss. MSE Loss reduced from 0.05...

This is the situation: Over 100 epochs: Train loss reduced from 0.2291 to 0.1962(almost plateau at around 25th epoch) Val loss reduced from 0.2033 to 0.1972(its not smooth and it...