RT DETRv2

🚀 Transformers - RT-DETRv2 Implementation

This project provides the Hugging Face (HF) transformers implementation for RT-DETRv2, offering an efficient solution for real - time object detection.

🚀 Quick Start

This is the HF transformers implementation for RT - DETRv2. The model used here is RT - DETRv2 - S.

RT - DETRv2 is an improved Real - Time DEtection TRansformer. It builds upon the previous state - of - the - art real - time detector, RT - DETR. It opens up a set of bag - of - freebies for flexibility and practicality, and optimizes the training strategy to achieve enhanced performance. To improve the flexibility, it is suggested to set a distinct number of sampling points for features at different scales in the deformable attention, enabling the decoder to achieve selective multi - scale feature extraction.

💻 Usage Examples

Basic Usage

import torch
import requests

from PIL import Image
from transformers import RTDetrForObjectDetection, RTDetrImageProcessor

url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
image = Image.open(requests.get(url, stream=True).raw)

image_processor = RTDetrImageProcessor.from_pretrained("jadechoghari/RT-DETRv2")
model = RTDetrForObjectDetection.from_pretrained("jadechoghari/RT-DETRv2")

inputs = image_processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)

results = image_processor.post_process_object_detection(outputs, target_sizes=torch.tensor([image.size[::-1]]), threshold=0.3)

for result in results:
    for score, label_id, box in zip(result["scores"], result["labels"], result["boxes"]):
        score, label = score.item(), label_id.item()
        box = [round(i, 2) for i in box.tolist()]
        print(f"{model.config.id2label[label]}: {score:.2f} {box}")