ViM-Disparity: Bridging the Gap of Speed, Accuracy and Memory for Disparity Map Generation
🤖 AI Summary
Addressing the longstanding trade-off among real-time performance, accuracy, and memory efficiency in stereo matching, this paper pioneers the integration of the Vision Mamba (ViM) architecture into disparity estimation, yielding a lightweight visual state space model. Methodologically, we synergize the long-range modeling capability of state space models (SSMs) with an end-to-end deep learning framework and introduce a multi-objective evaluation metric jointly quantifying inference latency, FLOPs, and accuracy. Experimental results demonstrate that our model achieves state-of-the-art accuracy on mainstream benchmarks—comparable to advanced CNN- and Transformer-based approaches—while significantly reducing GPU memory consumption by 37% and end-to-end latency to under 50 ms. This enables truly real-time, high-fidelity disparity map generation without compromising precision or computational efficiency.
Application Category
📝 Abstract
In this work we propose a Visual Mamba (ViM) based architecture, to dissolve the existing trade-off for real-time and accurate model with low computation overhead for disparity map generation (DMG). Moreover, we proposed a performance measure that can jointly evaluate the inference speed, computation overhead and the accurateness of a DMG model. The code implementation and corresponding models are available at: https://github.com/MBora/ViM-Disparity.
Problem
Research questions and friction points this paper is trying to address.
Real-time Disparity Generation
Computational Efficiency
Resource Management
Innovation
Methods, ideas, or system contributions that make the work stand out.
ViM-Disparity
Speed-Accuracy-Memory Balance
Comprehensive Evaluation Method
🔎 Similar Papers
No similar papers found.
