haizadtarik/object-detector

Text Guided or Image Guided Object Detection

This repository contains the code for implementing text-guided object detection and image-guided object detection.

The text-guided object detector is implmented using GroundingDINO and OWL-ViT from Huggingface Transformers.

The image-guided object detector is implemented using OWL-ViT from Huggingface Transformers and integrated with the text-guided object detector by using BLIP from Huggingface Transfomrers to get text prompt

Setup

1. Clone the repository

   git clone https://github.com/haizadtarik/object-detector.git
   

2. Install requirements

   pip install -r requirements.txt
   

3. Install and setup GroundingDino

   git clone https://github.com/IDEA-Research/GroundingDINO.git
   cd GroundingDINO/
   pip install -e .
   wget -q https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth
   cd ..
   

Example Usage

• Text Guided with GroundingDINO

  from detectors import textDetector
  import requests
  
  url = "http://images.cocodataset.org/val2017/000000039769.jpg"
  image_path = requests.get(query_url, stream=True).raw
  text = "cat"
  detector = textDetector("dino")
  boxes, logits, labels, image_target = detector.run(image_path=image_path, text_prompt=text, threshold=0.2)
  image_target.show()
  

• Text Guided with OWL-ViT

  from detectors import textDetector
  import requests
  
  url = "http://images.cocodataset.org/val2017/000000039769.jpg"
  image_path = requests.get(query_url, stream=True).raw
  text = "cat"
  detector = textDetector("owl")
  boxes, logits, labels, image_target = detector.run(image_path=image_path, text_prompt=text, threshold=0.2)
  image_target.show()
  

• Image Guided with OWL-ViT

  from detectors import imageDetector
  import requests
  
  url = "http://images.cocodataset.org/val2017/000000039769.jpg"
  image_path = requests.get(query_url, stream=True).raw
  
  query_url = "http://images.cocodataset.org/val2017/000000524280.jpg"
  query_image_path = Image.open(requests.get(query_url, stream=True).raw)
  
  detector = imageDetector()
  boxes, logits, labels, image_target = detector.run(image_path=image_path, query_image_path=query_image_path)
  image_target.show()
  

• Image guided with blip and dino

  from detectors import imageDetector
  import requests
  
  url = "http://images.cocodataset.org/val2017/000000039769.jpg"
  target_image_path = requests.get(query_url, stream=True).raw
  
  query_url = "http://images.cocodataset.org/val2017/000000524280.jpg"
  query_image_path = Image.open(requests.get(query_url, stream=True).raw)
  
  detector = imageDetector()
  boxes, logits, labels, image_target = detector.run(target_image_path, query_image_path, threshold=0.2, model_name="dino")
  image_target.show()