DDStereo: Efficient Dual Decoder Transformers for Stereo 3D Road Anomaly Detection

A new stereo vision model called DDStereo achieves state-of-the-art accuracy in 3D road anomaly detection while matching the real-time speed of monocular systems, according to research submitted to arXiv on June 23, 2026 [1][2]. The paper, titled "DDStereo: Efficient Dual Decoder Transformers for Stereo 3D Road Anomaly Detection," addresses two persistent challenges in stereo-based 3D object detection: real-time performance and open-set generalization [1][2]. Existing stereo methods typically achieve twice the accuracy of monocular approaches but suffer from significantly lower inference speeds, making them unsuitable for real-time applications [2]. DDStereo introduces a Dual-Decoder Stereo Transformer architecture with two lightweight decoder branches. One branch handles open-set foreground 2D detection, while the other performs 3D attribute regression. The decoders share object-level queries to achieve unified target-level alignment [1][2]. The researchers also designed a compact disparity feature extractor and a streamlined decoder architecture to enhance inference efficiency [2]. Experiments on public stereo 3D benchmarks showed that DDStereo achieves state-of-the-art accuracy under both closed-set and open-set protocols [1][2]. The method surpasses existing stereo 3D detectors in inference speed and, for the first time, achieves real-time performance comparable to monocular approaches [2]. The preprint was posted on arXiv, an open-access repository of electronic preprints that is moderated but not peer-reviewed [6]. As of November 2024, the repository receives about 24,000 submissions per month and has hosted over two million articles since its founding in 1991 [6]. The DDStereo paper appears under the Computer Vision and Pattern Recognition category [1]. The abstract page for the paper includes links to several community-developed tools through arXivLabs, a framework that allows collaborators to develop and share features directly on the arXiv website [4][5]. These tools include the Bibliographic Explorer, which displays citation relationships, and the CORE Recommender, which facilitates exploration of related open-access papers from a global network of research repositories [4][5]. arXivLabs partners must adhere to arXiv's values of openness, community, excellence, and user data privacy, and are granted only minimal and anonymized user data access [4].