InspectFlow

InspectFlow runs built-in conveyor scenes through an OpenCV inspection pipeline and replays the annotated frames in the browser. Parts keep a track ID as they move through the frame, and the station records pass or reject decisions at the inspection gate.

The paired view runs nominal and poor-light conditions with the same inspection settings so their results can be compared directly.

Run locally

Python 3.10 or newer is required.

cd inspect-flow
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt
python3 run.py

Open http://127.0.0.1:8000.

Use the station

• Play, pause, scrub, or change replay speed.
• Adjust the input condition and inspection thresholds, then select Run inspection.
• Select Compare conditions to run nominal and poor-light scenes and switch between the two completed replays.
• Read the current frame measurements, confirmed gate totals, defect counts, and paired result table.
• Download the decision log.

The gate sits at 25% of the frame width. A track receives one decision after it is observed before the gate and then crosses from left to right. A track that disappears early or first appears beyond the gate remains undecided.

HTTP API

Interactive API documentation is available at http://127.0.0.1:8000/docs.

Method and path	Purpose
GET /	Inspection station
GET /api/meta	Scenarios, defaults, limits, and inspection fields
POST /api/runs	Generate and inspect a run
GET /api/runs/{run_id}/frames/{index}.jpg	Annotated replay frame
GET /api/runs/{run_id}/inspection-log.csv	Gate decisions and measurements

Example:

curl -sS -X POST http://127.0.0.1:8000/api/runs \
  -H 'content-type: application/json' \
  -d '{"scenario":"nominal","frame_count":96,"config":{}}'

Runs are held in memory and expire as newer runs replace them. The server binds to loopback by default and has no authentication or TLS, so it should not be exposed to an untrusted network.

The thresholds are tuned for the included top-down scenes. Real cameras need their own lighting, lens, color, and dimensional calibration.