search

Chanuk-Yang / Deep_Continuous_Fusion_for_Multi-Sensor_3D_Object_Detection

32 stars 3 forks source link

the generation of BEV map #10

Open Machine97 opened 3 years ago

Machine97 commented 3 years ago

Hi Yang, It seems that the way of BEV map generation in the file data_import_carla.py is different from the paper. For example, in the paper, BEV input is generated by voxelizing the 3D space into a 512 × 448 × 32 volume, corresponding to 70 meters in front direction, ±40 meters on left and right sides of the ego-car. But in your code, the size of volume is 700 × 700 × 32, and the calibration matrix between Lidar and camera is not got from KITTI dataset. Otherwise, I don't know why the size of uv is 640 × 480. Could you please explain these questions for me. Thank you very much!

Chanuk-Yang commented 3 years ago

We use the customed carla dataset first. we didnt use KITTI dataset yet, but will be used for validate our algorithm. We will show the results with KITTI dataset when we make total algorithm.