This folder contains code to reproduce the results on end-to-end training of PIXOR model on stereo 3D object detection.
- pytorch 1.1.0 (torchvision 0.3.0)
- torch_scatter
- configargparse
- opencv-python
- tqdm
- numpy
- scipy
Download pixor model pretrained on SDN on KITTI here and pretrained depth model here.
compile the kitti offical evaluation code:
cd evaluation/kitti_eval
./compile.sh
run end2end finetuning (on 2 TITAN RTX GPUs):
CUDA_VISIBLE_DEVICES=0,1 python train_pixor_e2e.py -c configs/fusion.cfg --depth_pretrain <depth_model_pretrain_path> --pixor_pretrain <detection_model_pretrain_path>
CUDA_VISIBLE_DEVICES=0,1 python train_pixor_e2e.py -c configs/fusion.cfg --mode eval --eval_ckpt <full_model_checkpoint_path>