Siddhartha

If you are using torch>1.5, then you need to replace AT_CHECK with AT_ASSERT. See https://github.com/facebookresearch/votenet/issues/97

@AlfredCP Did your previous PR resolve this or does this need more work ?

@ElenaStrashko If you are using python, you can create channel name for camera image ` builder.stream("/camera/0").category(xviz_avs.CATEGORY.PRIMITIVE).type(xviz_avs.PRIMITIVE_TYPES.IMAGE)` Next you can stream it. Please note that `img`must shouldn't be a numpy uint8...

@CDitzel You may want to check out [this issue](https://github.com/karpathy/deep-vector-quantization/issues/7). The gumbel softmax function expects logits as inputs, so passing distances mayn't be appropriate. The scale of distances will be quite...

@enhuiz The standard Gumbel distribution i.e. G(0, 1) looks like this  When we call F.gumbel_softmax in pytorch, it adds the provided logits with G(0, 1) distribution. If the logits...

@CDitzel In [pytorch implementation](https://github.com/pytorch/pytorch/blob/4cb534f92ef6f5b2ec99109b0329f93a859ae831/torch/csrc/api/include/torch/nn/functional/activation.h#L177), they seem to be directly adding the logits to the sample from gumbel distribution. I believe they divide the logits by temperature before sampling from categorical...

I was trying to convert it to GPU but i came across the following issue.. 1. No return_counts in torch.unique. There are some hacky ways to solve this but it...

Thanks. For object part segmentation, does T-Net seem to improve performance (compared to network without it) ?

Yes, this feature will be extremely useful! I hope this gets added.

This issue has been fixed. Perhaps you are not using the latest commit ? Please try `pip install git+git://github.com/KieranWynn/pyquaternion` Then you can pass your rtol and atol values like this...