inference-cli with batch processing for Workflows

[PawelPeczek-Roboflow]

Description

New command in inference-cli to process data with Workflows, making it possible to:

• process videos - one passes video path and specifies output dir and can run any workflow, responses from each frame will be saved in output directory as CSV of JSONL file

• process individual image - results dumped into output dir - one may use python package or API to run workflow

• process directory of images - results dumped into output dir - with sub-dir for each image - one may use python package or API to run workflow and request aggregation of all JSON results into single JSONL or CSV

• when workflow returns image:

  • for video processing it saves output video file (only dimensionality=1 outputs, as otherwise it does not make sense)
  • for image processing - output is saved in image results sub-directory
  • one may disable image-outputs saving
  • image outputs do not get collected in JSON / CSV files to save space

• processing of directory of images is indempotent:

  • can resume after failure (processed images got denoted in log file)
  • this behaviour can be overriden - to force re-processing

• processing can run really fast against hosted API:

  • single-stage workflow with yolov8n-640 - one client could run over 65 images per second
  • 2-stage-workflow with yolov8n-640 and classifier - one client could run at 30 images per second

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

How has this change been tested, please provide a testcase or example of how you tested the change?

• CI still 🟢
• new tests

Any specific deployment considerations

For example, documentation changes, usability, usage/costs, secrets, etc.

Docs

• Docs updated? What were the changes: