Label Studio with YOLOv7 Backend

Label Studio ML Backend using YOLOv7 (https://github.com/yhsmiley/yolov7).

Currently only able to predict. Training has not been incorporated yet.

Quick Start

git clone https://github.com/tehwenyi/label-studio-backend-yolov7.git
cd label-studio-backend-yolov7/

Copy your YOLOv7 weights (.pt file) into the weights folder (create it if it doesn't exist) and rename it to weights.pt.

Please make sure that your weights have been reparameterised and the state dictionary has been saved. Refer to this link to find out how to reparameterise and save the state dictionary.

Copy your YOLOv7 deploy cfg (eg. cfg/deploy/yolov7.yaml) into the cfg folder (create it if it doesn't exist) and rename it to cfg.yaml.
Start running the backend with Docker Compose.

docker compose up

Follow the instructions on https://github.com/tehwenyi/label-studio to set up with Docker Compose.

In the Label Studio UI, open the project that you want to use with your ML backend.
Make sure that the labels you input in the project correspond exactly to your model's trained labels, otherwise you might get a No Label annotation after prediction.
Click Settings > Machine Learning.
Click Add Model.
Type a Title for the model and provide type http://server:9090 for the URL for the ML backend.
(Optional) Type a description.
(Optional) Select Retrieve predictions when loading a task automatically to allow predictions on all images.
Click Validate and Save. Instructions adapted from Label Studio Documentation

For any changes to the YOLOv7 inference parameters, edit the parameters in Line 21 of backend.py
The Docker containers are added to the same network label-studio.
The YOLOv7 backend uses Python 3.9.
The YOLOv7 backend uses port 9090 while label studio typically uses 8080.
To find out how to create your own ML backend, refer to heartexlabs/label-studio-ml-backend# label-studio-backend-yolov7