DY Zhang

@kellen5l Hi, I just use the pretrained file provided by this repo. Maybe you can try to set the batch_size=8 when testing your model. BTW, my environment ``` Ubuntu 16.04.7...

@kellen5l It seems that the IA-SSD doesn't really use the spconv lib, I don't think it is the problem of spconv version. In ```IA-SSD.yaml line 33 & 34 & 38```,...

@hhhmrcscs With the same weight file, all modules including the BNs should have same params, so the testing batch size should not affect the performance. Can you give me a...

@kellen5l I think you are right. The implementation of ```centroid-aware sampling``` is in ```IASSD_head.py```, I guess. The ```sa_centerness_mask``` corresponds to the mask used in the ```centroid-aware sampling``` module. ```python def...

To my understanding, when we treat the batch size as B * W, the "global CLS attention" is actually calculated within each window. If we use ViT backbones with both...

To my understanding, the attention maps calculated by the window/global attention are "local"/"global", which may have different meanings and properties. Can we use reshaping operations alone to transform them?

I understand, thank you!

Sorry for the late reply. I didn't encounter such errors before. You can go to the official repo of mmdetection3d for help.

You can have a try. Since nuScenes data is captured by 32-beam LiDAR while Waymo data is captured by 64-beam LiDAR, the point cloud of nuScenes is relatively more sparse...

> > I had same issue. > > I find the issue. The code does not save the weights of the diffusion model untrained. I had a similar issue here....