About generating imput depth

[shibing15]

Hi! When I process the FastLIVO dataset, I find that the projected point cloud is not as dense as shown in the reference image. Could you guide me on how to generate such a dense depth map? However, accumulating point clouds from multiple frames seems to be affected by pose errors, and the results are not ideal.

[jianhengLiu]

the point cloud image is the snapshot from rviz. The original fastlivo2's rviz should give yoi what you need

…

---Original--- From: "Yunda ***@***.***> Date: Mon, May 12, 2025 21:44 PM To: ***@***.***>; Cc: ***@***.***>; Subject: [hku-mars/M2Mapping] About generating imput depth (Issue #14) shibing15 created an issue (hku-mars/M2Mapping#14) Screenshot.from.2025-05-12.21-33-43.png (view on web) Hi! When I process the FastLIVO dataset, I find that the projected point cloud is not as dense as shown in the reference image. Could you guide me on how to generate such a dense depth map? However, accumulating point clouds from multiple frames seems to be affected by pose errors, and the results are not ideal. 000002.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[shibing15]

Sorry to bother you. I obtained the global map all_raw_points.pcd using the modified fastlivo2. Then I transformed the point cloud into the camera coordinate system using T_C_W @ pcd. However, I found that no points were projected onto the camera.

[shibing15]

Additionally, I also tried projecting the accumulated point clouds from the parsered_dataset onto the image. For each frame, I used color_pose_inv @ depth_pose @ pc, but the projection showed noticeable misalignment, as shown below. I checked the code and found that color_poses.txt and depth_poses.txt store the poses T_W_C and T_W_L respectively. My goal is to obtain the LiDAR depth map corresponding to each image. I would greatly appreciate your help! Thank you!

[jianhengLiu]

It seems that there is a mismatch between the sensor pose and the pointclouds. The 'depth' directory stores the pointclouds in LiDAR coordinates, FYI.
Or the first few frames might have an unstable pose estimation, which might lead to this misalignment. Check if there is also a misalignment in the middle frame.