Related Experiment Videos
LANet: A Lightweight and Accurate Balanced Network Based on State Space Models for Real-Time Semantic Segmentation
Summary
This study introduces LANet, a lightweight semantic segmentation model balancing accuracy and speed for real-time applications like autonomous driving. LANet achieves high performance with fewer parameters and faster inference, ideal for resource-constrained devices.
Area of Science:
- Computer Vision
- Deep Learning
- Artificial Intelligence
Background:
- Real-time semantic segmentation is vital for applications like autonomous driving.
- Current deep learning models often trade accuracy for speed or vice versa.
- A balance between high accuracy and fast inference is crucial for real-time semantic segmentation.
Purpose of the Study:
- To propose a novel lightweight semantic segmentation model, LANet, for real-time applications.
- To design a model that effectively balances segmentation accuracy and inference speed.
- To optimize performance for devices with limited computational resources.
Main Methods:
- Developed a lightweight and accurate balanced network (LANet).
- Integrated standard and depthwise convolutions.
- Introduced a multiskip concat bottleneck (MSC-B), pyramidal dual attention (FEPDA) modules, and multiscale state space models (FFMSSM) for feature enhancement and fusion.
Main Results:
- LANet significantly improves inference speed and reduces model parameters.
- Maintained high segmentation accuracy.
- Achieved 76.9% mIoU at 130.0 FPS with 1.33M parameters on an RTX 3090 GPU (without pretraining).
Conclusions:
- LANet offers a superior balance between accuracy and speed for real-time semantic segmentation.
- The model is well-suited for deployment on resource-limited platforms.
- LANet represents a significant advancement in efficient deep learning for computer vision tasks.
