Detectron v2 - Instance Segmentation

Instance Segmentation using Detectron v2

What is Detectron v2?

Detectron2 is Facebook AI Research's next generation software system that implements state-of-the-art object detection algorithms. This implementation results in a pipeline that will be able to take an image as an input and give the segmented image as a result.

Detectron v2 (October 2019) is a modification of the Detectron (jan 2019). The goal of the detectron was to create an hub of the implementation of the the differnt architectures like ResNeXt{50,101,152}, VGG16 etc. along with the diifferent algorithms like Mask RCNN, FPN etc.

Detectron v2 implements the new architectures along with new State of the art algorithms to provide an easy understanding.

How to use?

The downloaded file contain one file

./detectronv2.sh

If you want to test you model in isolated enviornment, you can use anaconda https://www.anaconda.com/distribution/

To create new enviornment conda create -n test python=3.6

conda activate test

Configuring Enviornment

The provided script will configure the enviornment for you. Two folder named detectron_model and cocoapi should be created after successful execution of the script

Steps to run the script: chmod 777 ./detectron.sh

./detectron.sh

Inference Steps

After running these steps, you will get the segmented image of the input image

wget http://images.cocodataset.org/val2017/000000439715.jpg -O input1.jpg

python demo/demo.py --config-file configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml \ --input input1.jpg \ --opts MODEL.WEIGHTS detectron2://COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl

you can pass multiple images also eg:--input input1.jpg input2.jpg

Dataset Used	COCO
Framework	PyTorch
OS Used	Linux
Publication	Detectron v2