SQLFlow Object Detection

About The Project

This project contains a modified step Image to perform Object Detection on SQLFlow using TensorFlow.

Project Structure

• datasets/: this directory contains a small set of images and its annotations based on the PASCAL VOC dataset.
• step/: the code to be called from a SQFLow client. It also contains the labels and a pre-trained model based on ResNet50.
• Dockerfile: defines the Docker image.
• requirements.txt: dependencies for the step image.

Usage

Build the Docker Image by running:

docker build -t <your-registry>/<image-name>:<tag> -f Dockerfile .

Push it to your own image registry:

docker push <your-registry>/<image-name>:<tag>

Call it from an extended SQLFlow query, using the TO RUN clause.

%%sqlflow
DROP TABLE IF EXISTS voc.result;
SELECT * FROM voc.annotations
TO RUN <your-registry>/<image-name>:<tag>
CMD "tf_object_detection.py",
    "--dataset=voc"
INTO result;

Resources

• SQLFlow extends SQL to support AI.
• The PASCAL Visual Object Classes dataset.
• Tensorflow Object Detection API.

License

Apache License 2.0