Image segmentation ML model deployed as a RESTful API and packaged in a Docker container.
To build the API I used Flask and Flasgger with SwaggerUI. Flasgger allows us to automatically generate an interactive API documentation where one can try out API calls.
The semantic segmentation model deployed is a Fully-Convolutional Network model with a ResNet-50 backbone, pre-trained on the COCO train2017 dataset and containing the same classes as Pascal VOC (20).
Running the API from a Docker container:
- From the api folder, build the docker container image with:
$ docker image build -f Dockerfile.pytorch -t <image-name> .- Run the container from the image created:
$ docker container run -d --rm -p 8888:8888 <image-name>- Go to http://localhost:8888/apidocs/ to interact with the API. You can test POST calls with the sample images in this repository or uploading your own images.
api.mp4
Visualizing responses:
- To visualize the outputted segmented image you can download the Response body (it downloads as a .txt file), and then pass it to the function response2img that you can find in utils.py.
