Implementation of "A CNN-Based Online Self-Calibration of Binocular Stereo Cameras for Pose Change"
paper link
Google drive link gives the pre-trained weights of calibnet, lccnet, zhang's network, and our network.
PSMNet's KITTI2012 weights are used for evaluation.
google drive link
PSMNet link
Download the kitti odometry dataset.
Replace all calib.txt in data folder with calib.txt downloaded from "Download odometry data set (calibration files, 1 MB)".
(('Tr') line does not exist, giving an error.)
Open the cfg file in the CFG folder and modify the paths.
python >= 3.10.9. I worked in an anaconda environment.
pytorch >= 1.13.1 (https://pytorch.org/)
opencv (pip install opencv-python)
easydict (pip install easydict)
numba (pip install numba)
Check out the readme.md in /scripts
I lost the settings I used for the experiment.
Results using the code provided by this repository will not match the results of the paper.
| method | Mean (paper) | Std. Dev (paper) | SSIM (paper) | Mean (github) | Std. Dev (github) | SSIM (github) | |
|---|---|---|---|---|---|---|---|
| calibnet | 1.44 | 2.87 | 0.9885 | 1.19(-0.25) | 2.89(+0.02) | 0.9893(+0.0008) | |
| lccnet | 2.52 | 5.05 | 0.9698 | 2.12(-0.40) | 4.44(-0.61) | 0.9740(+0.0042) | |
| zhang | 0.49 | 0.89 | 0.9970 | 0.71(+0.43) | 1.72(+0.83) | 0.9944(-0.0026) | |
| ours | 0.21 | 0.71 | 0.9985 | 0.27(+0.06) | 1.00(+0.29) | 0.9980(-0.0005) |
There are personal information in the GY dataset. (Vehicle registration plate, identifiable face)
Therefore, I cannot in any way provide experiments on the GY dataset.
LCCNet: Lidar and Camera Self-Calibration Using Cost Volume Network