search

ZhaochongAn / COSeg

[CVPR 2024] This repo contains the code for our paper: Rethinking Few-shot 3D Point Cloud Semantic Segmentation
49 stars 4 forks source link

Scannet dataset #8

Open Lzy-dot opened 2 days ago

Lzy-dot commented 2 days ago

Do I need to download the entire Scannet dataset?

ZhaochongAn commented 2 days ago

We provide Preprocessed Datasets, which you can download directly for training and testing without needing to process the entire dataset yourself.

For visualization, we utilize the full scenes as input to the model. In the first step of visualization, we set data_root to [PATH_to_DATASET_processed_data]/scenes/data instead of [PATH_to_DATASET_processed_data]/blocks_bs1_s1/data. The scenes/data can be obtained after completing the second step in the data preprocessing process. I will update the Preprocessed Datasets to include the results of scenes/data so that you can directly access them by downloading as well!

Please also note that the current visualization is only tested on the S3DIS dataset. Extending it to ScanNet might require some adjustments. Feel free to submit pull requests if you achieve this!

Lzy-dot commented 2 days ago

Thank you!

ZhaochongAn commented 2 days ago

I have updated the Preprocessed Datasets to include scenes/data. You should now be able to download them directly without needing to do the preprocessing yourself.

AmingWu commented 2 days ago

@Lzy-dot , do you perform visualization successfully?