angelsantamaria/yolopv2

YOLOPv2: Better, Faster, Stronger for Panoptic driving Perception

YOLOPv2:rocket:: Better, Faster, Stronger for Panoptic driving Perception

Cheng Han, Qichao Zhao, Shuyi Zhang, Yinzi Chen, Zhenlin Zhang, Jinwei Yuan

[YOLOPv2 arXiv Preprint]

News

• August 30, 2022: We've released the inference code / trained model and published web demo, just enjoy it !

• August 24, 2022: We've released the tech report for YOLOPv2. This work is still in progress and code/models are coming soon. Please stay tuned! ☕️

Introduction

😁We present an excellent multi-task network based on YOLOP💙,which is called YOLOPv2: Better, Faster, Stronger for Panoptic driving Perception. The advantages of YOLOPv2 can be summaried as below:

• Better👏: we proposed the end-to-end perception network which possess better feature extraction backbone, better bag-of-freebies were developed for dealing with the training process.
• Faster✈️: we employed more efficient ELAN structures to achieve reasonable memory allocation for our model.
• Stronger💪: the proposed model has stable network design and has powerful robustness for adapting to various scenarios .

Results

We used the BDD100K as our datasets,and experiments are run on NVIDIA TESLA V100.

Web Demo

• Integrated into Huggingface Spaces 🤗 using Gradio. Try out the Web Demo !

Visualization

model : trained on the BDD100k dataset and test on T3CAIC camera.

Model parameter and inference speed

Model	Size	Params	Speed (fps)
YOLOP	640	7.9M	49
HybridNets	640	12.8M	28
YOLOPv2	640	38.9M	91 (+42) ⏫

Traffic Object Detection Result

Result	Visualization
Model [email protected] (%) Recall (%) MultiNet 60.2 81.3 DLT-Net 68.4 89.4 Faster R-CNN 55.6 77.2 YOLOv5s 77.2 86.8 YOLOP 76.5 89.2 HybridNets 77.3 92.8 YOLOPv2 83.4(+6.1)⏫ 91.1(-1.7) ⬇️

Drivable Area Segmentation

Result	Visualization
Model Drivable mIoU (%) ——:relaxed:—— MultiNet 71.6 DLT-Net 71.3 PSPNet 89.6 YOLOP 91.5 HybridNets 90.5 YOLOPv2 93.2(+1.7) ⬆️

Lane Line Detection

Result	Visualization
Model Accuracy (%) Lane Line IoU (%) Enet 34.12 14.64 SCNN 35.79 15.84 Enet-SAD 36.56 16.02 YOLOP 70.5 26.2 HybridNets 85.4 31.6 YOLOPv2 87.3(+1.9)⬆️ 27.2(-4.4) ⬇️

Day-time and Night-time visualization results

Models

You can get the model from here.

Demo Test

We provide two testing method.You can store the image or video.

python demo.py  --source data/example.jpg

Third Parties Resource

• YOLOPv2 NCNN C++ Demo: YOLOPv2-ncnn from FeiGeChuanShu
• YOLOPv2 ONNX and OpenCV DNN Demo: yolopv2-opencv-onnxrun-cpp-py from hpc203

License

YOLOPv2 is released under the MIT Licence.