senzhang-github / ekf-imu-depth Goto Github PK

Hi @chaimi2013 @SenZHANG-GitHub, thank you for open-sourcing your work!

Would you like to release your pretrained weights for reproduction? Thanks!

I have RTK data and image data，and I have matched RTK data to the image data。Can I train the network on my custom data。
Wait for your help.

Hi,
The download links are expired or not accessible anymore.

Is there a way you can guide me to generate the dataset myself?

Hello teacher, do you have any relevant organization methods for datasets? At present, I am facing several problems:

1. Is the path to the dataset a root path?
2. Can the organization of the dataset provide relevant examples?
3. The dataset includes those parts.

I have been testing you model some, and I was wondering about this trans_scale_factor which is set to 5.4 by default.

From my understanding, this has to do with the stereo-configuration in KITTI. As such we should not use it when running with mono only? That is: set trans_scale_factor = 1.0 when not training using stereo.
Otherwise the relative motions between images will be wrong. This wouldn't be an issue if we only used the pose-network since it could learn to deal with this scale. However, this scale is also used in the compute_imu_pose_with_inv function, meaning that the acceleration-based translation will be incorrect. (Though this may be somewhat corrected by the velocity network)

Have I understood this correctly? I saw that my predicted depths were surprisingly low before I changed this scale factor to 1.

Thanks for your work!
Using IMU is very instructive for solving the scale problem of monocular depth estimation.

I think providing a detailed description of data preparation can help others better follow your research.

Hello. Could you provide the files after executing match_kitti_imu.py?

Dear author,
We want to train the model in other datasets that contain the IMU data. How we can define the Trans_scale_factor in the training? Trans_scale_factor is depth scale?

Hello, can the final result predict the absolute depth?

It is commendable that your work has been very meticulous, but if you can list the organization format of each file in your Kitti dataset, it will be more friendly to beginners~