Xiaoyang Wu
Xiaoyang Wu
Oh, are you using PTv3? We introduce a structure call "Point" in PTv3, and since PTv3, backbone return Point instead of feature. I will optimize this later.
Haha, save some time. I am majorly working on pre-training and backbone. I only need a simple baseline as PointGroup to help me verify the effectiveness of these things. Hope...
Sorry, for super late reponse. I am working on it now.
Alright, solved here: https://github.com/Pointcept/Pointcept/commit/58e101fe17c11527e587f5476bdd0685667e3dfa
Maybe currently, you can train from scratch or use weight from our PPT+PTv3. We will launch a very strong SSL pre-trained model at the end of this year. It is...
Nice! Good to hear that.
Our community usually uses Area 5 for evaluation. For 6-fold validation, you have to change the config to set val and test to each area and use our script to...
Hey, here is the link for the config for the data pipeline for the validation split (https://github.com/Pointcept/Pointcept/blob/main/configs/nuscenes/semseg-pt-v3m1-0-base.py#L133-L149). Could you first align the data process pipeline with the above validation pipeline?...
Hi, the preprocessing code for ScanNet200 is the same as for ScanNet20. The code is designed for both benchmarks. https://github.com/Pointcept/Pointcept/blob/main/pointcept/datasets/preprocessing/scannet/preprocess_scannet.py#L159
The following code should be helpful: https://github.com/Pointcept/Pointcept/commit/bebf24bef0f3730b0c570fae080479bbbc8b68f9 https://github.com/Pointcept/Concerto/blob/main/demo/0_pca_outdoor.py