ifcb-classifier-wrapper

Dockerfile for WHOI's ifcb_classifier, with wrapper to facilitate bulk classification.

Building

To build and test run

docker build -t ifcb-classifier-wrapper .
docker run --rm ifcb-classifier-wrapper --help

which should produce

Usage: run.py [OPTIONS] IFCB_DATA_DIR OUTPUT_DIR IFCB_CLASSIFY_MODEL_PATH

Options:
  --start-date [%Y-%m-%d]
  --end-date [%Y-%m-%d]
  --since-days INTEGER
  --run-id TEXT
  --config FILE
  --force, --clobber / --no-force
  --pid TEXT
  --help                          Show this message and exit.

Running

Example classification of last week of data:

docker run --rm --gpus all --name ifcb-recent-classify \
	-e CUDA_VISIBLE_DEVICES='0' \
	-v /path/to/ifcb/raw:/data/ifcb/raw:ro \
	-v /path/to/ifcb/model:/data/ifcb/model:ro \
	-v /path/to/ifcb/classified:/data/ifcb/classified \
	ifcb-classifier \
	--since-days 7 \
  /data/ifcb/raw \
  /data/ifcb/classified \
  /data/ifcb/model/themodel.ptl

For a full backfill, omit --since-days. For a specify date range, omit --since-days and provide --start-date and --end-date instead (YYYY-MM-DD). To force generation of existing class files, set --force.

More usage details can be found in the ifcb_classifier wiki.