vennsa/classfier

Simple YOLO vehicle-count server

Two endpoints (upload image → get counts):

• POST /count_total returns total vehicles only.
• POST /count_by_type returns car/motorcycle/bus/truck + total.
• POST /priority_decision accepts two images and returns which lane has priority + gate open time.

Setup

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Run

python app.py

Example request (PowerShell)

$img = "C:\path\to\frame.jpg"

Invoke-RestMethod -Method Post `
  -Uri "http://127.0.0.1:8000/count_by_type" `
  -Form @{ image = Get-Item $img }

Example request (2 images, priority)

$a = "C:\path\to\laneA.jpg"
$b = "C:\path\to\laneB.jpg"

Invoke-RestMethod -Method Post `
  -Uri "http://127.0.0.1:8000/priority_decision" `
  -Form @{ image_a = Get-Item $a; image_b = Get-Item $b }

Notes

• Vehicle classes are: car, motorcycle, bus, truck (COCO ids: 2, 3, 5, 7).
• The first run will download a YOLO model (yolov8n.pt) automatically.
• Set detection threshold with CONF (default 0.25) and model with YOLO_MODEL.
• Gate time policy is configurable with CAR_SEC, MOTORCYCLE_SEC, BUS_SEC, TRUCK_SEC, plus MIN_GATE_SEC and MAX_GATE_SEC.
• If you have custom models for winner/time, place them in Model/models/:
  • winner_classifier.pkl
  • gate_seconds_regressor.pkl