Selected work

Representative Papers

Cross-Stage Attention Propagation for Efficient Semantic Segmentation

Apr 07, 2026

This work addresses the significant computational redundancy in existing lightweight semantic segmentation methods, which independently compute attention at each level of multi-scale decoders. To overcome this inefficiency, the authors propose a Cross-Stage Attention Propagation (CSAP) mechanism that computes attention only once at the deepest feature layer and efficiently propagates it to shallower layers, thereby eliminating redundant query-key operations while preserving multi-scale contextual modeling capability. Built upon CSAP, the lightweight model CSAP-Tiny achieves 42.9% mIoU on ADE20K with only 5.5 GFLOPs, outperforming SegNeXt-Tiny by 1.8% mIoU while reducing computational cost by 16.8%. This approach is the first to enable cross-stage sharing of attention distributions, effectively balancing efficiency and performance.

0 citationsRead paper

Language-Grounded Multi-Domain Image Translation via Semantic Difference Guidance

Jan 12, 2026

Existing methods struggle to accurately map the semantic differences conveyed by natural language prompts into corresponding visual changes in multi-domain image translation while preserving irrelevant content. To address this, this work proposes a semantic difference-guided mechanism that explicitly decomposes the semantic discrepancy between source and target prompts to generate attribute-level translation vectors, enabling fine-grained and composable cross-domain control. We design a GLIP-Adapter to integrate global semantics with local structural features and introduce a multi-domain control guidance mechanism that supports independent intensity modulation for each attribute and cross-modal alignment. Experiments on CelebA(Dialog) and BDD100K demonstrate that our approach outperforms existing baselines in terms of visual fidelity, structural consistency, and interpretability of domain-specific control.

0 citationsRead paper

Neural ATTF: A Scalable Solution to Lifelong Multi-Agent Path Planning

Apr 21, 2025

Existing approaches to lifelong multi-agent path finding (MAPF) in dynamic warehouse logistics suffer from critical bottlenecks in scalability, real-time adaptability, and planning efficiency. To address these challenges, this paper proposes the Adaptive Task Token Framework (ATTF), the first method to jointly optimize latency-sensitive task scheduling and collision-aware path planning by integrating Priority-Guided Task Matching (PGTM) with a data-driven Neural Space-Time A* (STA*) algorithm. ATTF incorporates reinforcement learning–inspired heuristics, spatiotemporal A* search, dynamic priority scheduling, and online task reassignment. Evaluated on standard benchmarks, ATTF significantly outperforms state-of-the-art methods—including TPTS, CENTRAL, and LNS-wPBS—achieving a 32% increase in system throughput, a 47% reduction in average planning latency, and enabling millisecond-scale coordinated planning for over one hundred agents.

0 citationsRead paper

Retrieval-Augmented Fine-Tuning With Preference Optimization For Visual Program Generation

Feb 23, 2025

Low generation accuracy of ladder diagram (LD) visual programming language in industrial automation, compounded by the inadequacy of existing prompting methods for complex domain-specific configurations. Method: We propose a two-stage training paradigm: (1) retrieval-augmented fine-tuning (RAFT) to enhance generalization by leveraging high-frequency subroutine patterns in industrial LDs; and (2) graph-editing–based direct preference optimization (DPO), which automatically constructs semantically consistent preference pairs to mitigate annotation scarcity and configuration complexity. Contribution/Results: This work is the first to synergistically integrate RAFT and DPO for visual program generation and introduces the first graph-editing–driven preference construction method tailored to LDs. On a real-world LD dataset, our approach achieves over 10% higher program-level accuracy than supervised fine-tuning, and even small models outperform large-model prompting methods—significantly improving reliability and practicality of industrial code generation.

0 citationsRead paper

Recent publications

Latest Papers

Cross-Stage Attention Propagation for Efficient Semantic Segmentation

Apr 07, 2026

This work addresses the significant computational redundancy in existing lightweight semantic segmentation methods, which independently compute attention at each level of multi-scale decoders. To overcome this inefficiency, the authors propose a Cross-Stage Attention Propagation (CSAP) mechanism that computes attention only once at the deepest feature layer and efficiently propagates it to shallower layers, thereby eliminating redundant query-key operations while preserving multi-scale contextual modeling capability. Built upon CSAP, the lightweight model CSAP-Tiny achieves 42.9% mIoU on ADE20K with only 5.5 GFLOPs, outperforming SegNeXt-Tiny by 1.8% mIoU while reducing computational cost by 16.8%. This approach is the first to enable cross-stage sharing of attention distributions, effectively balancing efficiency and performance.

0 citationsRead paper

Language-Grounded Multi-Domain Image Translation via Semantic Difference Guidance

Jan 12, 2026

Existing methods struggle to accurately map the semantic differences conveyed by natural language prompts into corresponding visual changes in multi-domain image translation while preserving irrelevant content. To address this, this work proposes a semantic difference-guided mechanism that explicitly decomposes the semantic discrepancy between source and target prompts to generate attribute-level translation vectors, enabling fine-grained and composable cross-domain control. We design a GLIP-Adapter to integrate global semantics with local structural features and introduce a multi-domain control guidance mechanism that supports independent intensity modulation for each attribute and cross-modal alignment. Experiments on CelebA(Dialog) and BDD100K demonstrate that our approach outperforms existing baselines in terms of visual fidelity, structural consistency, and interpretability of domain-specific control.

0 citationsRead paper

Neural ATTF: A Scalable Solution to Lifelong Multi-Agent Path Planning

Apr 21, 2025

Existing approaches to lifelong multi-agent path finding (MAPF) in dynamic warehouse logistics suffer from critical bottlenecks in scalability, real-time adaptability, and planning efficiency. To address these challenges, this paper proposes the Adaptive Task Token Framework (ATTF), the first method to jointly optimize latency-sensitive task scheduling and collision-aware path planning by integrating Priority-Guided Task Matching (PGTM) with a data-driven Neural Space-Time A* (STA*) algorithm. ATTF incorporates reinforcement learning–inspired heuristics, spatiotemporal A* search, dynamic priority scheduling, and online task reassignment. Evaluated on standard benchmarks, ATTF significantly outperforms state-of-the-art methods—including TPTS, CENTRAL, and LNS-wPBS—achieving a 32% increase in system throughput, a 47% reduction in average planning latency, and enabling millisecond-scale coordinated planning for over one hundred agents.

0 citationsRead paper

Retrieval-Augmented Fine-Tuning With Preference Optimization For Visual Program Generation

Feb 23, 2025

Low generation accuracy of ladder diagram (LD) visual programming language in industrial automation, compounded by the inadequacy of existing prompting methods for complex domain-specific configurations. Method: We propose a two-stage training paradigm: (1) retrieval-augmented fine-tuning (RAFT) to enhance generalization by leveraging high-frequency subroutine patterns in industrial LDs; and (2) graph-editing–based direct preference optimization (DPO), which automatically constructs semantically consistent preference pairs to mitigate annotation scarcity and configuration complexity. Contribution/Results: This work is the first to synergistically integrate RAFT and DPO for visual program generation and introduces the first graph-editing–driven preference construction method tailored to LDs. On a real-world LD dataset, our approach achieves over 10% higher program-level accuracy than supervised fine-tuning, and even small models outperform large-model prompting methods—significantly improving reliability and practicality of industrial code generation.

0 citationsRead paper