search

AmrElsersy / Stereo-3D-Detection

Stereo based 3D object detection on KITTI dataset using Pytorch implementing the Pseudo LIDAR pipeline with papers: AnyNet & PointPillars & SFA3D
42 stars 11 forks source link

custom dataset #3

Open eungjoolee opened 2 years ago

eungjoolee commented 2 years ago

is this repo available to use it for custom dataset application?

AmrElsersy commented 2 years ago

mmmm yeah, but the depth network(AnyNet) is trained on KITTI Stereo dataset, and the 3d detection network (SFA3D) is trained on the generated dataset from the depth network

you can try the full pipeline on your custom stereo data, if you had a bad accuracy I think you will have to train both depth & 3d detection networks on your own data (the training code is available you just need to convert the format of your dataset to KITTI dataset format)

KexuanXia commented 3 months ago

mmmm yeah, but the depth network(AnyNet) is trained on KITTI Stereo dataset, and the 3d detection network (SFA3D) is trained on the generated dataset from the depth network

you can try the full pipeline on your custom stereo data, if you had a bad accuracy I think you will have to train both depth & 3d detection networks on your own data (the training code is available you just need to convert the format of your dataset to KITTI dataset format)

Hello, thank you for your excellent work. If I understand correctly, the SFA3D checkpoint you provided in the repo was trained on the pseudo lidar generated by AnyNet rather than real lidar data, right?