Skip to content

The official implementation of "CenterNeXt: Revisiting CenterNet in 2022"

License

Notifications You must be signed in to change notification settings

peternara/CenterNeXt

Repository files navigation

CenterNeXt

License: CC BY-NC 4.0

The official implementation of "CenterNeXt: Revisiting CenterNet in 2022"

It has not been published yet.

Results and Pre-trained Models

VOC2007

backbone resolution mAP FPS(on Titan Xp) FPS(on RTX 3090) FLOPs
(G)
model config weights
ResNet-18 512 x 512 74.92 110 164 14.8 config model
+ Coupled head 512 x 512 74.61 112 167 14.2 config model
+ Detection on lower resolution 512 x 512 74.36 127 191 13.0 config model
+ Mosaic augmentation 512 x 512 74.20 127 191 13.0 config model
+ Mixup augmentation 512 x 512 75.84 127 191 13.0 config model
ResNet-50 512 x 512 80.46 65 104 25.0 config model
ResNet-101 512 x 512 83.00 39 60 44.5 config model
ConvNeXt-T 512 x 512 83.57 43 87 26.8 config model

Setup

Create a new conda virtual environment

conda create -n centernext python=3.8 -y
conda activate centernext

Install Pytorch and torchvision following official instructions. For example:

pip install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio===0.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Clone this repo and install required packages:

git clone https://github.com/MarkAny-Vision-AI/CenterNeXt
cd CenterNeXt
pip install -r requirements.txt

Download a dataset:

if (your_os == 'Window'):

cd CenterNeXt
scripts/download-voc0712.bat
python create_gt_files.py

else:

cd CenterNeXt
scripts/download-voc0712.sh
python create_gt_files.py

Training

python train.py --model ./configs/models/your_model.yaml

Evaluation

python eval.py --model ./configs/models/your_model.yaml --weights /path/to/your_model.pth

Profiling

python profile.py --model ./configs/models/your_model.yaml

License

This project is licensed under the terms of the Attribution-NonCommercial 4.0 International license. It is released for academic research only and is free to researchers from educational or research institutes for non-commercial purposes.

Please see the LICENSE file for more information.

Please contact [email protected] or [email protected] for business inquiries.

About

The official implementation of "CenterNeXt: Revisiting CenterNet in 2022"

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Languages