search

Lab-of-AI-and-Robotics / GS_ICP_SLAM

[ECCV 2024] RGBD GS-ICP SLAM
MIT License
245 stars 26 forks source link

About Tracking On TUM Dataset #22

Open Buffyqsf opened 2 months ago

Buffyqsf commented 2 months ago

Thanks for your great work! I installed the pygicp library and tried to test it on TUM dataset and Replica dataset with the example you gave (fast_gicp/python_tester). And I found that the tracking accuracy of Replica dataset is similar to that of the article, but it is difficult to achieve the tracking accuracy of the article on TUM dataset. In your approach, tracking is achieved through pygicp.FastGICP (), but what other factors or tricks brings a better tracking accuracy?

Lightingooo commented 2 months ago

I have the same question.

Riboha commented 2 months ago

As mentioned in the paper, in the GS-ICP SLAM, we select keyframes by our dynamic selection method. But this method is not implemented in the python_tester of fast_gicp. And further details are added to the GS-ICP SLAM. Please checkout the paper for more details.