After I prepared the scannet200 dataset according to the process, some problems occurred during runtime
RobinLiuZX opened this issue · 2 comments
Since you did not provide the config.py files for the two initial datasets of scannet and scannet200 in the open source code, I tried to refer to the config files of scannet200 and scannet200-sv in oneformer3d and ESAM, but when running the train.py file, the following error was reported. At the same time, I found that the data in the scannet200 folder I got was missing the seg_info folder and the npy files under it. Could you help me solve this problem?
Our model is an online model that takes RGB-D video frame sequences as input, while the two initial datasets, scannet and scannet200, contain reconstructed point cloud data suitable for offline models like Oneformer3D. In our approach, these two initial reconstructed datasets are only used for calculating segmentation metrics. Additionally, we did not use the seg_info folder and the files under it; thank you for pointing this out. We will update the README of scannet and scannet200 to clarify our dataset processing.
This repo does not contain the code for offline perception. You can refer to OneFormer3D for training offline models.