telesto.ai Model API Template Project

This serves as a brief guide to help you submit your solutions for telesto.ai competitions. For a complete guide, see the technical documentation!

Preparing model archive

These instructions are for a classification model but for other types of models the steps are the same.

• Use the classification/model directory for preparing a model archive.
• Add weights of a model you developed to the directory.
• Add inference code. The model inference class should inherit telesto.models.ClassificationModelBase. For more information, see the telesto-base package, where the base classes are defined.
• Pack an archive with all needed files using cd classification; zip -r model.zip model.

A concrete and worked out example can be found here. For a detailed guide, see https://docs.telesto.ai!

Building Docker image and starting container

Building the image containing the model:

docker build --pull -t telestoai/model-api classification/model

Running the container:

docker run -p 9876:9876 --name model-api --rm telestoai/model-api

Testing classification model API

Send a sample input

curl -X POST -H "Content-Type:application/json" --data-binary @classification/example_data/example-input.json -i \
    http://localhost:9876/
...
{
    "predictions": [
        {"probs": {"cat": 0.32015, "dog": 0.67985}, "prediction": "dog"},
        {"probs": {"cat": 0.81545, "dog": 0.18455}, "prediction": "cat"}
    ]
}

Testing segmentation model API

Post a sample input

curl -X POST -H "Content-Type:application/json" --data-binary @segmentation/example_data/example-input.json -i \
    http://localhost:9876/jobs
...
{
    "job_id": "b741bd19767441f6b7abd022744083c9"
}

Get the result

curl -H "Content-Type:application/json" -i http://localhost:9876/jobs/b741bd19767441f6b7abd022744083c9
...
{
    "mask": {
        "content": "<BASE_64_IMAGE>"
    }
}

An example classification model

A fully worked out example of a classification model can be found in the classification/example_model directory. You can try it out just like your model as well, using the commands:

docker build --pull -t telestoai/model-api-example classification/model_example

docker run -p 9876:9876 --name model-api --rm telestoai/model-api-example

curl http://localhost:9876/

curl -X POST -H "Content-Type:application/json" --data-binary @classification/example_data/example-input.json -i \
    http://localhost:9876/

An example segmentation model

A complete segmentation example model can be prepared, using the following commands:

curl -o segmentation/model_example/model.pt https://telesto-ai-content.fra1.digitaloceanspaces.com/example-models/segmentation/model.pt

docker build --pull -t telestoai/model-api-example segmentation/model_example

docker run -p 9876:9876 --name model-api --rm telestoai/model-api-example

curl http://localhost:9876/

curl -X POST -H "Content-Type:application/json" --data-binary @segmentation/example_data/example-input.json -i \
    http://localhost:9876/jobs

curl -H "Content-Type:application/json" -i http://localhost:9876/jobs/<job_id>

Evaluating classification model running in container

pip install -r requirements.txt
python classification/evaluate.py --api-url http://localhost:9876 \
    --dataset-path classification/example_data/test_dataset --metric accuracy

Evaluating segmentation model running in container

pip install -r requirements.txt
python segmentation/evaluate.py --api-url http://localhost:9876 \
    --dataset-path segmentation/example_data/test_dataset --class-n 3