Chenhui Yang

Meituan LongCat Team, Anchun Gui, Bei Li et al.

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

Mohammed A. M. Elhassan, Chenhui Yang, Chenxi Huang et al.

Modern high-performance semantic segmentation methods employ a heavy backbone and dilated convolution to extract the relevant feature. Although extracting features with both contextual and semantic information is critical for the segmentation tasks, it brings a memory footprint and high computation cost for real-time applications. This paper presents a new model to achieve a trade-off between accuracy/speed for real-time road scene semantic segmentation. Specifically, we proposed a lightweight model named Scale-aware Strip Attention Guided Feature Pyramid Network (S$^2$-FPN). Our network consists of three main modules: Attention Pyramid Fusion (APF) module, Scale-aware Strip Attention Module (SSAM), and Global Feature Upsample (GFU) module. APF adopts an attention mechanisms to learn discriminative multi-scale features and help close the semantic gap between different levels. APF uses the scale-aware attention to encode global context with vertical stripping operation and models the long-range dependencies, which helps relate pixels with similar semantic label. In addition, APF employs channel-wise reweighting block (CRB) to emphasize the channel features. Finally, the decoder of S$^2$-FPN then adopts GFU, which is used to fuse features from APF and the encoder. Extensive experiments have been conducted on two challenging semantic segmentation benchmarks, which demonstrate that our approach achieves better accuracy/speed trade-off with different model settings. The proposed models have achieved a results of 76.2\%mIoU/87.3FPS, 77.4\%mIoU/67FPS, and 77.8\%mIoU/30.5FPS on Cityscapes dataset, and 69.6\%mIoU,71.0\% mIoU, and 74.2\% mIoU on Camvid dataset. The code for this work will be made available at \url{https://github.com/mohamedac29/S2-FPN

Mohammed A. M. Elhassan, Chenhui Yang, Chenxi Huang et al.

The following is a technical report to test the validity of the proposed Subspace Pyramid Fusion Module (SPFM) to capture multi-scale feature representations, which is more useful for semantic segmentation. In this investigation, we have proposed the Efficient Shuffle Attention Module(ESAM) to reconstruct the skip-connections paths by fusing multi-level global context features. Experimental results on two well-known semantic segmentation datasets, including Camvid and Cityscapes, show the effectiveness of our proposed method.

Meituan LongCat Team, Anchun Gui, Bei Li et al.

We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5% (from 19, 653 to 6, 965) on AIME-25, without degrading task accuracy. We release LongCat-Flash-Thinking to promote further advances in reasoning systems and agentic AI research.

Zongliang Zhang, Hongbin Zeng, Jonathan Li et al.

This paper deals with the geometric multi-model fitting from noisy, unstructured point set data (e.g., laser scanned point clouds). We formulate multi-model fitting problem as a sequential decision making process. We then use a deep reinforcement learning algorithm to learn the optimal decisions towards the best fitting result. In this paper, we have compared our method against the state-of-the-art on simulated data. The results demonstrated that our approach significantly reduced the number of fitting iterations.

Shanxin Zhang, Cheng Wang, Zhuang Yang et al.

The timely provision of traffic sign information to drivers is essential for the drivers to respond, to ensure safe driving, and to avoid traffic accidents in a timely manner. We proposed a timely visual recognizability quantitative evaluation method for traffic signs in large-scale transportation environments. To achieve this goal, we first address the concept of a visibility field to reflect the visible distribution of three-dimensional (3D) space and construct a traffic sign Visibility Evaluation Model (VEM) to measure the traffic sign visibility for a given viewpoint. Then, based on the VEM, we proposed the concept of the Visual Recognizability Field (VRF) to reflect the visual recognizability distribution in 3D space and established a Visual Recognizability Evaluation Model (VREM) to measure a traffic sign visual recognizability for a given viewpoint. Next, we proposed a Traffic Sign Timely Visual Recognizability Evaluation Model (TSTVREM) by combining VREM, the actual maximum continuous visual recognizable distance, and traffic big data to measure a traffic sign visual recognizability in different lanes. Finally, we presented an automatic algorithm to implement the TSTVREM model through traffic sign and road marking detection and classification, traffic sign environment point cloud segmentation, viewpoints calculation, and TSTVREM model realization. The performance of our method for traffic sign timely visual recognizability evaluation is tested on three road point clouds acquired by a mobile laser scanning system (RIEGL VMX-450) according to Road Traffic Signs and Markings (GB 5768-1999 in China), showing that our method is feasible and efficient.

Yan Yan, Hanzi Wang, Cuihua Li et al.

In this paper, an effective unconstrained correlation filter called Uncon- strained Optimal Origin Tradeoff Filter (UOOTF) is presented and applied to robust face recognition. Compared with the conventional correlation filters in Class-dependence Feature Analysis (CFA), UOOTF improves the overall performance for unseen patterns by removing the hard constraints on the origin correlation outputs during the filter design. To handle non-linearly separable distributions between different classes, we further develop a non- linear extension of UOOTF based on the kernel technique. The kernel ex- tension of UOOTF allows for higher flexibility of the decision boundary due to a wider range of non-linearity properties. Experimental results demon- strate the effectiveness of the proposed unconstrained correlation filter and its kernelization in the task of face recognition.