Yolov9 + Clip = Open-Vocabulary Object Detection

Introduction

• This project is based on Yolov9, added with CLIP, to realize Open-Vocabulary Object Detection without retraining the Yolov9 model.
• From the repositories above, you can download the necessary codes, datasets, etc.
• This repository only provides the detect_ov.py and val_ov.py used for OV detection.

Quickstart

• run the instruction in the terminal

  python detect_ov.py --source './data/images/horses.jpg' --img 640 --device cpu --weights './yolov9-c-converted.pt' --name yolov9_ov
  python val_ov.py --data data/objects365_val128.yaml --img 640 --batch 32 --conf 0.001 --iou 0.7 --device cpu --weights './yolov9-c-converted.pt' --save-json --workers 4 --name ovdetection

Preview

• The results of OV-Detection

  

• The results of OV-Validation