Hugging Face's logo

Spaces:
NeerajCodz
/
ObjectDetection
Running

App Files Community
ObjectDetection / README.md
NeerajCodz's picture
NeerajCodz
Sync from GitHub
074874b
preview code
|
raw
history blame
5.65 kB

🚀 Object Detection with Transformer Models

This project provides an object detection system using state-of-the-art transformer models, such as DETR (DEtection TRansformer) and YOLOS (You Only Look One-level Series). The system can detect objects from uploaded images or image URLs, and it supports different models for detection and segmentation tasks. It includes a Gradio-based web interface and a FastAPI-based API for programmatic access.

You can try the demo online on Hugging Face: Demo Link.

Models Supported

The following models are supported, as defined in the application:

  • DETR (DEtection TRansformer):

    • facebook/detr-resnet-50: DETR with ResNet-50 backbone for object detection. Fast and accurate for general use.
    • facebook/detr-resnet-101: DETR with ResNet-101 backbone for object detection. More accurate but slower than ResNet-50.
    • facebook/detr-resnet-50-panoptic(currently has bugs): DETR with ResNet-50 for panoptic segmentation. Detects objects and segments scenes.
    • facebook/detr-resnet-101-panoptic(currently has bugs): DETR with ResNet-101 for panoptic segmentation. High accuracy for complex scenes.

  • YOLOS (You Only Look One-level Series):

    • hustvl/yolos-tiny: YOLOS Tiny model. Lightweight and fast, ideal for resource-constrained environments.
    • hustvl/yolos-base: YOLOS Base model. Balances speed and accuracy for object detection.

Features

  • Image Upload: Upload images from your device for object detection via the Gradio interface.
  • URL Input: Input an image URL for detection through the Gradio interface or API.
  • Model Selection: Choose between DETR and YOLOS models for detection or panoptic segmentation.
  • Object Detection: Detects objects and highlights them with bounding boxes and confidence scores.
  • Panoptic Segmentation: Some models (e.g., DETR panoptic variants) support detailed scene segmentation with colored masks.
  • Image Properties: Displays image metadata such as format, size, aspect ratio, file size, and color statistics.
  • API Access: Use the FastAPI endpoint /detect to programmatically process images and retrieve detection results.

How to Use

1. Normal Git Clone Method

Follow these steps to set up the application locally:

Prerequisites

  • Python 3.8 or higher
  • Install dependencies using pip

Clone the Repository 

git clone https://github.com/NeerajCodz/ObjectDetection.git
cd ObjectDetection

Install Dependencies

Install the required dependencies from requirements.txt:

pip install -r requirements.txt

Run the Application

Start the FastAPI server using uvicorn:

uvicorn objectdetection:app --reload

Alternatively, launch the Gradio interface by running the main script:

python app.py

Access the Application

  • For FastAPI: Open your browser and navigate to http://localhost:8000 to use the API or view the Swagger UI.
  • For Gradio: The Gradio interface URL will be displayed in the console (typically http://127.0.0.1:7860).

2. Running with Docker

If you prefer to use Docker to set up and run the application, follow these steps:

Prerequisites

  • Docker installed on your machine. If you don’t have Docker, download and install it from here.

Download the docker Image

First, Pull the docker Image:

docker pull neerajcodz/objectdetection:latest

Run the Docker Container

Once the image is built, run the application using this command:

docker run -d -p 8080:80 neerajcodz/objectdetection:latest

This will start the application on port 8080. Open your browser and go to http://localhost:8080 to access the interface.

3. Demo

You can try the demo directly online through Hugging Face's Spaces:

Object Detection Demo

Using the API (Instable)

You can interact with the application via the FastAPI /detect endpoint to send images and get detection results.

Endpoint: /detect

POST: /detect

Parameters:

  • file: (optional) Image file (must be of type image/*).
  • image_url: (optional) URL of the image.
  • model_name: (optional) Choose from facebook/detr-resnet-50, hustvl/yolos-tiny, etc.

Example Request Body:

{
  "image_url": "https://example.com/image.jpg",
  "model_name": "facebook/detr-resnet-50"
}

Response:

The response includes a base64-encoded image with detections, detected objects, confidence scores, and unique objects with their scores.

{
  "image_url": "data:image/png;base64,...",
  "detected_objects": ["person", "car"],
  "confidence_scores": [0.95, 0.87],
  "unique_objects": ["person", "car"],
  "unique_confidence_scores": [0.95, 0.87]
}

Development Setup

If you'd like to contribute or modify the application:

  1. Clone the repository:
git clone https://github.com/NeerajCodz/ObjectDetection.git
cd ObjectDetection
  1. Install dependencies:
pip install -r requirements.txt
  1. Run the FastAPI server or Gradio interface:
uvicorn objectdetection:app --reload

or

python app.py
  1. Open your browser and navigate to http://localhost:8000 (FastAPI) or the Gradio URL (typically http://127.0.0.1:7860).

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests for bug fixes or new features on the GitHub repository.