search

TUMFTM / DPFT

Dual Perspective Fusion Transformer for Camera-Radar-based Object Detection
Apache License 2.0
27 stars 3 forks source link

where's 'calib_camera_lidar.txt' file #1

Closed YueSun0609 closed 1 month ago

YueSun0609 commented 3 months ago

Hi,

Thank you for your work! May I ask where I can find the 'calib_camera_lidar.txt' file? Since I downloaded the data from Google Driver and there's no 'calib_camera_lidar.txt' file inside 'info_calib' folder.

ffent commented 2 months ago

Hi @YueSun0609 ,

first of all, thanks for your interest in our work. It is correct, that the calib_camera_lidar.txt file is not included in the downloadable files from Google Drive. Therefore, we reached out to the original authors of the K-Radar dataset and they provided a lidar to camera calibration for us.

To fix this issue, just create a calib_camera_lidar.txt file under the data/kradar/raw/1/info_calib/ folder and insert the following values.

rx, rxy, rxz, tx, ryx, ry, ryz, ty, rzx, rzy, rz, tz
629.238090, -527.036820, -4.59938064, 1413.21630, 372.013478, 9.42970300, -559.685543, 1076.37845, 0.999925370, 0.0122165356, 0.000106612091, 1.84000000

However, as pointed out in our paper (Section 4.3) the calibration seems not to be perfect and calibration errors can be observed, which could have a negative impact on the performance. To counteract this, we have tried to fine-tune the provided calibration parameters by using the manual calibration tool provided by K-Radar, but it is very difficult to tune them manually. We also tried to use automated calibration tools from the OpenCalib toolbox but without any success. Ultimately, we have used the provided calibration parameters for all of our tests.

YueSun0609 commented 1 month ago

Hi @YueSun0609 ,

first of all, thanks for your interest in our work. It is correct, that the calib_camera_lidar.txt file is not included in the downloadable files from Google Drive. Therefore, we reached out to the original authors of the K-Radar dataset and they provided a lidar to camera calibration for us.

To fix this issue, just create a calib_camera_lidar.txt file under the data/kradar/raw/1/info_calib/ folder and insert the following values.

rx, rxy, rxz, tx, ryx, ry, ryz, ty, rzx, rzy, rz, tz
629.238090, -527.036820, -4.59938064, 1413.21630, 372.013478, 9.42970300, -559.685543, 1076.37845, 0.999925370, 0.0122165356, 0.000106612091, 1.84000000

However, as pointed out in our paper (Section 4.3) the calibration seems not to be perfect and calibration errors can be observed, which could have a negative impact on the performance. To counteract this, we have tried to fine-tune the provided calibration parameters by using the manual calibration tool provided by K-Radar, but it is very difficult to tune them manually. We also tried to use automated calibration tools from the OpenCalib toolbox but without any success. Ultimately, we have used the provided calibration parameters for all of our tests.

Got it! Thank you so much for your explanation!