stereo_rcnn_pytorch

This repository has been created to train stereo rcnn and convert pretrained model from pytorch to tensorflow using onnx library.

0.0. Install Pytorch:

conda create -n env_stereo python=3.6
conda activate env_stereo
conda install pytorch=1.0.0 cuda90 -c pytorch
conda install torchvision -c pytorch

0.1. Other dependencies:

git clone https://github.com/tomshalini/stereo_rcnn_pytorch.git
cd stereo_rcnn_pytorch
pip install -r requirements.txt

0.2. Build:

cd lib
python setup.py build develop
cd ..

Create symlinks:

cd data/kitti
ln -s stereo_rcnn_pytorch/object object
cd ../..

Training

Download the Res-101 pretrained weight https://drive.google.com/file/d/1_t8TtUevtMdnvZ2SoD7Ut8sS1adyCKTt/view, and put it into data/pretrained_model

Set corresponding CUDA_VISIBLE_DEVICES in train.sh, and simply run
It consumes ~11G GPU memery for training. 
The trained model and training log are saved in /models_stereo by default.

Testing

Download trained weight https://drive.google.com/uc?id=1rZ5AsMms7-oO-VfoNTAmBFOr8O2L0-xt&export=download and put it into models_stereo/, then just run
test_net.py

Inference

run predict.py