CSIRO Runner

Usage

Terminal 1

python poller.py \
  --watch_path=path/to/directory --output_file=path/to/output.csv \
  --batch_size=8

Terminal 2 (On the same machine)

The script currently recognizes the models from TF OD API, or the ones that have the same signature.

python serving.py \
  --model_path=path/to/saved_model --model_signature="serving_default"

To update the service proto python files, run

python -m grpc_tools.protoc -I.  --python_out=. --grpc_python_out=. service.proto

Poller

The poller watches for the *.jpg file additions in the specified directory. When new files are added, it will add them to dataset and fetch an image per second. This should be improved to match up with realtime performance requirement.

Detector

The detector runs inference, and maintains gRPC endpoint to accept inference requests and returning detection results. service.proto defines the gRPC service.

Benchmark

Simply measure the model's performance.

python benchmark.py \
  --model_path=path/to/model --image_path=path/to/image/dir --batch_size=1 \
  --model_signature="serving_default"

TODO

• Load management between poller and detector to achieve > 10fps performance.
• Dockerize (taeheej@)
• Tracker integration (swatisingh@)
• Performance optimizations (cheshire@)