Python 3.10 pytorch 2.6.0 docs
We've pre-trained YOLO-World-T/M from scratch and evaluate on the TT100K val , GTSDB and VOC.
We use TT100k and convert it to YOLO format to evaluate the SOTA of models
| model | Params | FLOPs | |||||
|---|---|---|---|---|---|---|---|
| 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 |
SMDDF is developed based on torch==2.6.0 and CUDA Version==11.8.
git clone https://github.com/rainbowyuyu/SMDDFNet.git
pip install -r requirements.txt
cd selective_scan && pip install . && cd .. pip install -v -e .
python SMDDF_train.py
- This repo is modified from open source real-time object detection codebase Ultralytics.
- The selective-scan from VMamba.
- The Mamba-backbone from Mamba-Yolo