Xiaoyang Wu

> ScanNet20 ``` [2022-12-30 11:44:06,249 INFO test.py line 108 2355837] Val result: mIoU/mAcc/allAcc 0.7547/0.8288/0.9115 [2022-12-30 11:44:06,249 INFO test.py line 110 2355837] Class_0 - wall Result: iou/accuracy 0.8609/0.9585 [2022-12-30 11:44:06,249 INFO...

The data structure is a bit changed in v1.5.2, you can refer changelog here: https://github.com/Pointcept/Pointcept/releases/tag/v1.5.2 ``` dataset |- split1 (e.g. train) |- data 1 |- coord.npy |- color.npy |- normal.npy...

Hi, use config for unScenes (https://github.com/Pointcept/Pointcept/blob/main/configs/nuscenes/semseg-pt-v3m1-0-base.py) as a base and set batch size to 1 and run the code with a single GPU for training latency and memory. And then,...

Good point! Set `pdnorm_adaptive=False` doesn't mean naive joint training, naive joint training causes negative transfer as discussed in the PPT paper. PDNorm consists of two components: 1. scale, bias adaptive...

> For memory, do you take peak memory? The code doesn't print memory usage, does it? I think you can simply achieve this by Google how to record the peak...

> Hi, thanks for your patience. I still have a simple question that should empty_cache set to True when measuring memory comsumption? Looking forward to your reply. @Charlie839242 I didn't...

Hi, try the below new InformationWriter. But not ensure the same as the PTv3 paper. In the PTv3 paper, I use a subprocess to track GPU memory usage. Recently, I...

Hi, any model on Pointcept supports point clouds with variable sizes. Check here: https://github.com/Pointcept/Pointcept?tab=readme-ov-file#offset.

Simply write a test hook which is the same as the validation process?

Yes, of course. Just modify the point group + spinet config with the ptv3 backbone config.