SMDDFNet: State-space Modeling and Dynamic Dual Fusion Network for Traffic Sign Detection

Model Zoo

We've pre-trained YOLO-World-T/M from scratch and evaluate on the TT100K val , GTSDB and VOC.

Dataset

We use TT100k and convert it to YOLO format to evaluate the SOTA of models

Inference on TT100K dataset

model	Params	FLOPs	${AP}_{{50}}^{val}$	${AP}_{{95}}^{val}$	${AP}_{{S}}^{val}$	${AP}_{{M}}^{val}$	${AP}_{{L}}^{val}$
SMDDF-T	6.1M	14.3G	80.4	60.1	50.4	64.5	79.6
SMDDF-M	21.8M	49.7G	87.7	68.2	58.1	75.0	83.5

Getting started

1. Installation

SMDDF is developed based on torch==2.6.0 and CUDA Version==11.8.

2.Clone Project

git clone https://github.com/rainbowyuyu/SMDDFNet.git

3. Install torch

pip install -r requirements.txt

4. Install selective_scan

cd selective_scan && pip install . && cd ..
pip install -v -e .

7. Training SMDDFNet

python SMDDF_train.py

Acknowledgement

• This repo is modified from open source real-time object detection codebase Ultralytics.
• The selective-scan from VMamba.
• The Mamba-backbone from Mamba-Yolo