Closed 2311762665 closed 2 years ago
As mentioned in the paper, we used the rendered 2D images only during the pre-training stage. All the downstream tasks are performed on the pre-trained point cloud feature extractor.
During pre-training, we used ShapeNet data and the corresponding 2D images are obtained from [1] as mentioned in the paper. Can you let me know which part of the notebook is incomprehensible ? We used the same notebook to produce the reported results in the paper.
[1] Qiangeng Xu, Weiyue Wang, Duygu Ceylan, Radomir Mech, and Ulrich Neumann. DISN: Deep implicit surface network for high-quality single-view 3D reconstruction. In Advances in Neural Information Processing Systems, volume 32, 2019
I tried to open the eval_ssl.ipynb file inside jupyter with no problem. Also, please tell me which version you used when you did the ScanObjectNN experiment, and the partition should be ‘trainning’, right?
Sorry for the delay in response.
We use the standard same version of ScanObjectNN as the previous works used. Kindly refer to [1] for more details. The partition should 'training' to fit the linear classifier and evaluated in the 'testing' partition.
[1] Hanchen Wang, Qi Liu, Xiangyu Yue, Joan Lasenby, and Matthew J. Kusner. Unsupervised point cloud pre-training via occlusion completion. In International Conference on Computer Vision, ICCV, 2021
Hello, dear author! How did you get the 2D images corresponding to the ModelNet40, ScanObjectNN point cloud data? The content inside eval_ssl.ipynb looks incomprehensible, can you provide the original .py file code?