Sharpiless

Question about size loss.

1

In function loss_size, the "gt_box_sizes" is scaled by point_cloud_dims on the scale of 0 to ∞ while the "pred_box_sizes" is scaled into (0, 1) by sigmoid. F.l1_loss is used to...

This may be a problem with incomplete uploaded code when running scripts/sunrgbd_ep1080.sh. This same as line4 in scripts/sunrgbd_ep1080.sh without a "\\".

Using random input yields even higher accuracy (95.77901%) and faster convergence in toy experiment😄 Of course it's not the same in higher dimensions, but I'm just going to mention this...

about Instance-Level Feature Denoising

2

关于FPN_Tensorflow实例级去噪模块，一条通道卷积得到的W*H*(C+1)这个特征图，我阅读源码时发现这个mask仅用于loss计算而未用于推理，想请教一下这个问题，，，（或者说图中Guide体现在哪里呢？） 

关于teacher model预训练权重

1

我发现您并未给出yolov5s在VOC上训练的权重，因此将无法复现README中的实验，请问能否上传一下呢

关于deepsort

1

好像没看到deepsort

Unable to find 'org.pytorch:pytorch_android:1.8.0-SNAPSHOT'

2

I find that 'org.pytorch:pytorch_android:1.8.0-SNAPSHOT' was deleted in https://oss.sonatype.org/#nexus-search;quick~pytorch_android，and there are compiling failure of other versions or APP flashback.

RuntimeError: The size of tensor a (129) must match the size of tensor b (132) at non-singleton dimension 2. Thanks for your contributions. But I have no idea what's wrong....

MMCV version mismatch

4

The [mmcv-nwd](https://github.com/jwwangchn/mmcv-nwd.git) provided is 1.2.7 while this repo requires 1.3.0+.

Thank you very much for your contribution, but the hyperlink of the pre-trained model has a 404 error. Could you please re-publish it? Grateful!