Arrow Research search

Author name cluster

Sijia Peng

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers
2 author rows

Possible papers

2

EAAI Journal 2026 Journal Article

Progressive feature learning framework with alternate fusion for object tracking based on visible and thermal infrared images

  • Xiaoyu Ni
  • Sijia Peng
  • Liang Yuan
  • Rihan Hai
  • Yadong Wang
  • Yongming Han

Object tracking based on the fusion of visible light and thermal infrared has attracted increasing attention because two modalities provide complementary information for robust tracking in challenging environments. However, how to effectively exploit multimodal context and promote sufficient interaction between modality-specific and modality-shared information remains a key challenge. In this work, we propose a progressive feature learning framework based on a vision Transformer backbone for object tracking using visible light and thermal infrared fusion. The proposed method introduces a spatial feature adaptive learning fusion module to produce fused multimodal features by learning spatially varying modality weights, and a cross-fusion discrimination module to enhance target–background discrimination by refining fused template and search representations. Different from conventional progressive fusion pipelines that mainly emphasize repeated modality fusion, our framework follows an alternating progressive learning strategy that repeatedly switches between modality-complementary fusion and tracking-oriented discrimination refinement across network depth. In this way, modality-specific information and modality-shared information are collaboratively optimized at different semantic levels and stages. Extensive experiments on standard tracking benchmarks demonstrate that the proposed method achieves competitive accuracy and efficiency compared with recent advanced trackers. These results indicate that the proposed method is suitable for artificial intelligence applications requiring reliable object tracking in real-world scenarios under low illumination, partial occlusion, and adverse conditions, such as intelligent surveillance, transportation monitoring, unmanned system perception, search and rescue, and safety-critical observation.

ICML Conference 2025 Conference Paper

Rethinking Time Encoding via Learnable Transformation Functions

  • Xi Chen 0072
  • Yateng Tang
  • Jiarong Xu
  • Jiawei Zhang 0001
  • Siwei Zhang 0001
  • Sijia Peng
  • Xuehao Zheng
  • Yun Xiong

Effectively modeling time information and incorporating it into applications or models involving chronologically occurring events is crucial. Real-world scenarios often involve diverse and complex time patterns, which pose significant challenges for time encoding methods. While previous methods focus on capturing time patterns, many rely on specific inductive biases, such as using trigonometric functions to model periodicity. This narrow focus on single-pattern modeling makes them less effective in handling the diversity and complexities of real-world time patterns. In this paper, we investigate to improve the existing commonly used time encoding methods and introduce Learnable Transformation-based Generalized Time Encoding (LeTE). We propose using deep function learning techniques to parameterize nonlinear transformations in time encoding, making them learnable and capable of modeling generalized time patterns, including diverse and complex temporal dynamics. By enabling learnable transformations, LeTE encompasses previous methods as specific cases and allows seamless integration into a wide range of tasks. Through extensive experiments across diverse domains, we demonstrate the versatility and effectiveness of LeTE.

v2026.09.13