uhvbyvtv/dresscode-access-control

ME-661 Vision-Based Dress-Code Access Control

This Hugging Face Space demonstrates a YOLO-based intelligent entrance access-control system.

Goal

Classify attire as:

• —INAPPROPRIATE if shorts or slippers/open footwear are detected
• —APPROPRIATE if no violation is detected

Then simulate:

• —Inappropriate → audible beep + door locked
• —Appropriate → door open

Model behavior

The app supports three model modes:

1. 1.Fine-tuned custom YOLO model

Upload your trained YOLO weights to:

   models/best.pt

1. 1.Pretrained YOLO-World fallback

If models/best.pt is missing, the app uses:

   yolov8s-worldv2.pt

This is a zero-shot YOLO-based model that can attempt to detect prompts like person, shorts, slippers, sandals, and open footwear.

1. 1.Plain YOLO11 fallback

If forced with MODEL_MODE=yolo11, the app loads:

   yolo11s.pt

Plain COCO YOLO11 can detect person, but it is not trained for shorts or slippers.

Recommended production setup

For best accuracy, train/fine-tune YOLO on your dress-code dataset and upload:

models/best.pt

The app will automatically use it.