rt-detr-v2

RT-DETR: DETRs Beat YOLOs on Real-time Object Detection

Input

(Image from http://www.crowdhuman.org/)

Ailia input shape : (1, 3, IMAGE_HEIGHT, IMAGE_WIDTH)

Output

Ailia output shape : (1, 3, IMAGE_HEIGHT, IMAGE_WIDTH)

Usage

Automatically downloads the onnx and prototxt files when running. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 rt-detr-v2.py

If you want to specify the input image, put the image path after the --input option.
You can use --savepath option to change the name of the output file to save.

$ python3 rt-detr-v2.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

By adding the --arch option, you can specify architecture type which is selected from "rtdetrv2_r18vd_120e","rtdetrv2_r34vd_120e","rtdetrv2_r50vd_6x","rtdetrv2_r50vd_m_7x","rtdetrv2_r101vd_6x" , (default: rtdetrv2_r18vd_120e)

$ python3 rt-detr-v2.py --arch rtdetrv2_r18vd_120e

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 rt-detr-v2.py --video VIDEO_PATH

Reference

RT-DETR

Framework

Pytorch

Model Format

ONNX opset=17

Netron

rtdetrv2_r18vd_120e_coco.onnx.prototxt

rtdetrv2_r34vd_120e_coco.onnx.prototxt

rtdetrv2_r50vd_6x_coco.onnx.prototxt

rtdetrv2_r50vd_m_7x_coco.onnx.prototxt

rtdetrv2_r101vd_6x_coco.onnx.prototxt