Detectron2

Mask RCNN, Pointrend, Faster RCNN and evaluation

Dependencies:

pip3 install -r requirements.txt

pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/{CUDA_VERSION}/{TORCH_VERSION}/index.html

git clone --branch v0.6 https://github.com/facebookresearch/detectron2.git detectron2_repo

pip install -e detectron2_repo  (If have to compile from source) (optional)

Version used locally (Cuda 11.7) - 
python3 -m pip install detectron2==0.6 -f   https://dl.fbaipublicfiles.com/detectron2/wheels/cu102/torch1.10/index.html

Steps: Download Kitti_official_semantic from the list of datasets in the shared folder on the system 10.165.81.147 change path names in code wherever necessary

run dependencies based on your system

python3 Mask_rcnn.py             To Run mask rcnn model through detectron2
python3 faster_rcnn.py           To Run faster rcnn model through detectron2
python3 masks.py                 To run and get masks for predictions and ground truth  (preds and gt)
python3 masks2.py                To run and get masks for predictions and ground truth for pedestrians and vehicles (grey scale images)
python3 pointrend.py             To Run pointrend image segmentation model through detectron2
python3 ptrend_masks.py          To generate pointrend masks and download masks numpy in .npy files
python3 ensemble.py              To get the ensemble of mask rcnn and pointrend masks stored as .npy files