WinMamba: Multi-Scale Shifted Windows in State Space Model for 3D Object Detection

2025-11-17

AI Summary

In 3D object detection, existing Mamba-based approaches are constrained by axis-aligned, fixed-size window scanning, compromising both computational efficiency and long-range spatial modeling. To address this, we propose WinMamba—a novel backbone network featuring: (1) a multi-scale sliding-window mechanism that relaxes axis-aligned constraints and enlarges receptive field coverage; (2) a window-scale adaptive module and learnable positional encoding to enhance geometric awareness; and (3) stacked WinMamba blocks integrating voxel-level feature compensation with multi-scale fusion. Evaluated on KITTI and Waymo Open Dataset, WinMamba achieves significant improvements over state-of-the-art baselines. Ablation studies confirm the effectiveness of each component. The code will be made publicly available.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning with Manifolds
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

3D object detection is critical for autonomous driving, yet it remains fundamentally challenging to simultaneously maximize computational efficiency and capture long-range spatial dependencies. We observed that Mamba-based models, with their linear state-space design, capture long-range dependencies at lower cost, offering a promising balance between efficiency and accuracy. However, existing methods rely on axis-aligned scanning within a fixed window, inevitably discarding spatial information. To address this problem, we propose WinMamba, a novel Mamba-based 3D feature-encoding backbone composed of stacked WinMamba blocks. To enhance the backbone with robust multi-scale representation, the WinMamba block incorporates a window-scale-adaptive module that compensates voxel features across varying resolutions during sampling. Meanwhile, to obtain rich contextual cues within the linear state space, we equip the WinMamba layer with a learnable positional encoding and a window-shift strategy. Extensive experiments on the KITTI and Waymo datasets demonstrate that WinMamba significantly outperforms the baseline. Ablation studies further validate the individual contributions of the WSF and AWF modules in improving detection accuracy. The code will be made publicly available.

Problem

• Balancing computational efficiency with long-range spatial dependency capture
• Addressing spatial information loss in fixed-window axis-aligned scanning
• Enhancing 3D object detection through multi-scale representation and contextual cues

Innovation

• Multi-scale shifted windows in state space model
• Window-scale-adaptive module for voxel feature compensation
• Learnable positional encoding with window-shift strategy