Implement model inference

Depends #2

The plan is to print the predicted polygons on an image from Arkindex to stdout.

Step 1: Model embedding

• Download model weights in models/, store though git-lfs
• Create a new method on the worker in worker.py, that loads this model and prepare a Predictor (DefaultPredictor from detectron2?)
• Call this method in the configure method to setup the predictor

Step 2: Image downloading and processing

• In process_element, call open_image
• Pass to predictor and retrieve the predictions

Step 3: Post-processing

• If it's not hard, you can save the predictions as PNG image in os.path.join(self.work_dir,"{element.id}.png"). This way, we could see it on Arkindex.
• Convert the predictions to a list of dict with the following structure

[
    {
        "type": ...,         # Channel/Class name of the object
        "polygon": ...,      # Anything that can identify the object, bounding box positions, full list of coordinates,...
        "confidence": ...,   # Confidence score
    },
    ...
]

• Dump this list to stdout

added P2 label

assigned to @Nishanthartham

mentioned in issue #4

You should be able to test your worker via:

1. Environment setup

export ARKINDEX_API_URL=https://demo.arkindex.org
export ARKINDEX_API_TOKEN=...
pip install -e .

2. Worker exec

worker-palmira --dev --element <element_uuid> --config <configuration file if needed>

unassigned @tlavigne