Official Code for Cross-Modality Fusion Transformer for Multispectral Object Detection.
Multispectral Object Detection with Transformer and Yolov5
Multispectral image pairs can provide the combined information, making object detection applications more reliable and robust in the open world. To fully exploit the different modalities, we present a simple yet effective cross-modality feature fusion approach, named Cross-Modality Fusion Transformer (CFT) in this paper. Unlike prior CNNs-based works, guided by the Transformer scheme, our network learns long-range dependencies and integrates global contextual information in the feature extraction stage. More importantly, by leveraging the self attention of the Transformer, the network can naturally carry out simultaneous intra-modality and inter-modality fusion, and robustly capture the latent interactions between RGB and Thermal domains, thereby significantly improving the performance of multispectral object detection. Extensive experiments and ablation studies on multiple datasets demonstrate that our approach is effective and achieves state-of-the-art detection performance.
Night Scene
Day Scene
Python>=3.6.0 is required with all requirements.txt installed including PyTorch>=1.7 (The same as yolov5 https://github.com/ultralytics/yolov5 ).
git clone https://github.com/DocF/multispectral-object-detection
$ pip install -r requirements.txt-[FLIR] [Google Drive] [Baidu Drive] extraction code:qwer
A new aligned version.
-[LLVIP] download
-[VEDAI] download
You need to convert all annotations to YOLOv5 format.
Refer: https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data
yolov5 weights (pre-train)
-[yolov5s] google drive
-[yolov5m] google drive
-[yolov5l] google drive
-[yolov5x] google drive
CFT weights
-[LLVIP] google drive
-[FLIR] google drive
some example in data/multispectral/
some example in models/transformer/
note!!! we used xxxx_transfomerx3_dataset.yaml in our paper.
| Dataset | CFT | mAP50 | mAP75 | mAP |
|---|---|---|---|---|
| FLIR | 73.0 | 32.0 | 37.4 | |
| FLIR | ✔️ | 78.7 (Δ5.7) | 35.5 (Δ3.5) | 40.2 (Δ2.8) |
| LLVIP | 95.8 | 71.4 | 62.3 | |
| LLVIP | ✔️ | 97.5 (Δ1.7) | 72.9 (Δ1.5) | 63.6 (Δ1.3) |
| VEDAI | 79.7 | 47.7 | 46.8 | |
| VEDAI | ✔️ | 85.3 (Δ5.6) | 65.9(Δ18.2) | 56.0 (Δ9.2) |
Log Average Miss Rate
| Model | Log Average Miss Rate |
|---|---|
| YOLOv3-RGB | 37.70% |
| YOLOv3-IR | 17.73% |
| YOLOv5-RGB | 22.59% |
| YOLOv5-IR | 10.66% |
| Baseline(Ours) | 6.91% |
| CFT(Ours) | 5.40% |
Miss Rate - FPPI curve
yolov3 yolov5 对应train.py train_multi_modal.py test.py predict.py脚本
yolov6 yolov8 yolov9 yolov11对应train_v11.py train_multi_modal_v11.py test_v11.py predict_v11.py脚本
其中models/yolo中的Model对应单模态模型，models/yolotest中的Model对应多模态模型
第一种是add相加，如：yolov5s_fusion_add.yaml PC2f_MPF_yolov8.yaml
第二种是transformer CFT融合：yolov5s_fusion_transformer.yaml
在这个rep已经支持检测模型的多机多卡训练，使用torchrun的方式进行。如果需要检测nccl是否正常
可以使用test_dist.py这个脚本测试。