DeTR-LAB

Library of Detection with Transformer

Requirements

• We recommend you to use Anaconda to create a conda environment:

conda create -n detr python=3.6

• Then, activate the environment:

conda activate detr

• Requirements:

pip install -r requirements.txt 

We suggest that PyTorch should be higher than 1.9.0 and Torchvision should be higher than 0.10.3. At least, please make sure your torch is version 1.x.

This repository

In this repository, you can enjoy:

• DETR
• Anchor-DETR
• DINO

Experiments

DETR

I evaluate DETR by loading official pretrained weight.

Model	backbone	FPS 3090	FLOPs	Params	AP	Weight
DETR-R50	R-50	37	95.2 B	36.7 M	41.7	github
DETR-R50-DC5	R-50	20	162.1 B	36.7 M	43.0	github
DETR-R101	R-101	25	174.7 B	55.7 M	43.1	github
DETR-R101-DC5	R-101	14	241.6 B	55.7 M	44.3	github

Anchor DETR

I evaluate AnchorDETR by loading official pretrained weight.

Model	backbone	FPS 3090	FLOPs	Params	AP	Weight
Anchor-DETR-R50	R-50	37	97.0 B	30.7 M	42.1	github
Anchor-DETR-R50-DC5	R-50	20	154.0 B	30.7 M	44.2	github
Anchor-DETR-R101	R-101	21	176.5 B	49.7 M	43.5	github
Anchor-DETR-R101-DC5	R-101	16	233.5 B	49.7 M	45.2	github

Train

sh train.sh

You can change the configurations of train.sh, according to your own situation.

However, limited by my GPU, I can't verify whether this repository can reproduce the performance of the official DETR.

Test

python test.py -d coco \
               --cuda \
               -v detr_r50 \
               --weight path/to/weight \
               --root path/to/dataset/ \
               --show

Evaluation

python eval.py -d coco-val \
               --cuda \
               -v detr_r50 \
               --weight path/to/weight \
               --root path/to/dataset/

Demo

I have provide some images in data/demo/images/, so you can run following command to run a demo:

python demo.py --mode image \
               --path_to_img dataset/demo/images/ \
               -v detr_r50 \
               --cuda \
               --weight path/to/weight \
               --show

If you want run a demo of streaming video detection, you need to set --mode to video, and give the path to video --path_to_vid。

python demo.py --mode video \
               --path_to_img dataset/demo/videos/video_name \
               -v detr_r50 \
               --cuda \
               --weight path/to/weight \
               --show

If you want run video detection with your camera, you need to set --mode to camera。

python demo.py --mode camera \
               -v detr_r50 \
               --cuda \
               --weight path/to/weight \
               --show