Andrey Penkovskiy

You can generate the pattern online or using OpenCV and fill your `.yaml` based on the given examples.

Just use AprilGrid board with enough tags (more than 3) and make sure the `.yaml` file, which describes the pattern and is passed through `--target` when running calibration, contains proper...

I've experienced similar issues as yours. By adding gradient computation to the [GES](https://github.com/ajhamdi/ges-splatting) from `diff-gaussian-rasterization-w-pose`, I've been trying to estimate the camera poses of my training dataset given ground truth...

@YufengJin Yes, it would be interesting to discuss this further. I also have some concerns about the sensitivity of the loss to the rotation matrix. In the last experiments, I...

Also, I think different exposure times might impact the optimization. It's not the case for me, since I've been using synthetic data with fixed exposure times. In MonoGS they estimate...

> Can I ask how you isolated the translation optimization from the rotational one ? By setting the rotational state update `camera.cam_rot_delta.data` to zero right after `pose_optimizer.step()` . P. S.:...

Scale drift typical for monocular odometry

Got the same issue. Using the torch implementation of SuperPoint solved the problem. You can use [this](https://github.com/Eunchan24/omniglue) fork in order to test with this implementation of SuperPoint. Or use the...

All the accelerations are in the same frame; `dR` is used to transform accelerations.

ORB-SLAM doesn't construct dense maps; check densification approaches, like OpenMVS. (one camera is enough). ORB-SLAM3 can provide good enough poses to run OpenMVS (especially in Stereo-Inertial mode).