search

maudzung / Complex-YOLOv4-Pytorch

The PyTorch Implementation based on YOLOv4 of the paper: "Complex-YOLO: Real-time 3D Object Detection on Point Clouds"
https://arxiv.org/pdf/1803.06199.pdf
GNU General Public License v3.0
1.21k stars 260 forks source link

Is it possible to test on custom data without using Kitti Dataset #66

Open prsnkmr opened 1 year ago

prsnkmr commented 1 year ago

I want to run this model on a single point cloud, without using kitti dataset is it possible?.If it is possible please help me to test this model on point clouds

regenthor commented 1 year ago

I tried that too, but failed. Anyone know? @maudzung

aorait commented 11 months ago

Yes you can test it using a point cloud stored in a bin file. To do it cleanly you can modify the test.py and dataloader to only use Lidar data. A more hacky solution is to create another folder with the same directory structure as kitti, copy any kitti image and rename it to the same serial id as your custom point clouds id, put this id in the ImageSets/test.txt, update the configs.dataset_dir in the test.py and you should see some detections using this command python test.py --gpu_idx 0 --pretrained_path your_path/Model_complexer_yolo_epoch_200.pth --show_image. The results don't seem great though (at least for me since I don't use velodyne data).