Haolin Liu

@xXuHaiyang The desk is no longer None currently and I update it in the readme file. The quantitative results is run on the updated file, which is zip as 3D-FUTURE-watertight.zip....

Yes, they should be consistent. I would recommend to use split-filter since this filter some highly occluded samples. And the pretrained weight is from training on the split-filter set. I...

你好，文章中的注意力模块没有详细说明。实际操作很简单，使用目标物体的global feature, 再经过几层mlp，就能得到一个channel-wise的权重 (该权重维度跟hourglass网络抽取的feature map的channel维度一致)。然后再给feature map的每一个channel，都乘于一个不同的权重，从而达到最终被过滤的feature map。详细可以看代码里面的attentional module。

Hi, the SUNRGBD loader and demo will be released in a few days. However, it is difficult to evaluate quantitatively in SUNRGBD dataset, so I may release a visualize code...

@khushijpatel Within tomorrow.

@khushijpatel Hi, the SUNRGBD demo is already updated.

Hi, is it only happening if the bgocc.zip? I will exam it ASAP

I also got automatic cancel during the downloading, I will re-uplode a new bgocc file ASAP.

I already update the link for bgocc.zip, the updated file is not corrupted anymore, but I still found out that occasionally the download will be canceled while it is not...

Hi the pix3d model is actually only trained on pix3d.