search

Lab-of-AI-and-Robotics / GS_ICP_SLAM

[ECCV 2024] RGBD GS-ICP SLAM
MIT License
297 stars 34 forks source link

The problem of reduced frame continuity #29

Closed Tokyo-dream closed 3 months ago

Tokyo-dream commented 4 months ago

Thanks again for the excellent work However, through the experiment, the images in the Replica room0 data set are selected 1 frame every 5 frames to get 400 pictures, and GS-ICP is run, but the effect is not ideal. Would you like to ask whether this is the limitation of the method or what parameters can be adjusted to solve it. Compared to SplaTAM, the effect is normal SplaTAM

屏幕截图 2024-05-17 122348

GS-ICP-SLAM download

Riboha commented 3 months ago

Hello! In the original replica dataset, the movement of the camera between frames is small, and the default parameters of GS-ICP-SLAM are fitted to this environment. Could you try adjusting the max_correspondence_distance parameter to be a bit larger? (ex: 0.1)