p890040 / Understand-MaskRCNN

Understand-MaskRCNN is a extremely simple FasterRCNN, MaskRCNN repo for explaining how it works.

Understand-MaskRCNN

Understand-MaskRCNN is a extremely simple FasterRCNN, MaskRCNN repo for explaining how it works.

About The Project

• The main purpose of this project is creating a easy to understand code, because FasterRCNN/MaskRCNN is still the strongest basline in two-stage object detection algorithm so far. I try to avoid using extra packages, so that users don't need to install and take time to learn other works.

• Furthermore, I try to unlock every black box in the algorithm. It allows whether you are a beginner, half-understanding, or only unclear one or two points, you can fully understand it through this project.

• Note: If you want a powerful and efficient MaskRCNN tool. Please refer to Detectron2, maskrcnn-benchmark, mmdetection (Pytorch),and Tensorflow Object Detection API , matterport (Tensorflow)

Installation

pip install numpy
pip install opencv-python
pip install torch==1.4.0 torchvision==0.5.0

Getting Started

Train

Open MaskRCNN/main.py

data_path = os.path.join(os.path.dirname(os.getcwd()), 'dataset')
train_net(data_path, max_epoch = 300)

Inference

Open MaskRCNN/main.py

data_path = os.path.join(os.path.dirname(os.getcwd()), 'dataset')
inference(data_path, score_thresh = 0.8)

TODO

• Region Proposal Network
• Faster RCNN
• Mask RCNN
• The article of Understand-MaskRCNN
• Flexible anchors mask rcnn