Yikang Zhou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yikang Zhou has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Shunping Ji
Wuhan University
3 shared publications - Tao Zhang
Wuhan University
3 shared publications - Xiangtai Li
Nanyang Technological University, Singapore
3 shared publications - Haochen Wang
New Laboratory of Pattern Recognition, State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China
2 shared publications - Lu Qi
Wuhan University, China
2 shared publications Dizhe Zhang
Insta360
1 shared publication
Latest Publications
- 10 September 2026
- Published by Springer Nature
- p. 586-604
Actor as Its Own Critic: Unifying Region Understanding and Localization via CycleGRPO
- 10 September 2026
- Published by Springer Nature
- p. 586-604
Actor as Its Own Critic: Unifying Region Understanding and Localization via CycleGRPO
- Open Access
- 23 July 2026
- Published by MDPI AG in Preprints.org
Visual Tokenizers: A Comprehensive Survey
Visual tokenization bridges the gap between high-dimensional visual data and sequence modeling by transforming images, videos, and 3D content into compact token sequences. Recent advances in Multimodal Large Language Models (LLMs) have further underscored the critical role of visual tokenizers. Despite this rapid progress, the literature in this area remains fragmented. This work addresses this gap by presenting a comprehensive taxonomy of visual tokenizers. We review the evolution of task-specific tokenizers along two complementary dimensions: high-level tokenizers that emphasize semantic representations and low-level tokenizers that focus on pixel reconstruction and compression. Motivated by the emergence of Multimodal Large Language Models capable of supporting both understanding and generation tasks, we highlight recent advances in unified tokenizers. These models aim to integrate semantic abstraction with fine-grained visual details within a single representation. We categorize unified tokenizers into single-encoder and dual-encoder architectures and further organize them according to their semantic preservation strategies and fusion mechanisms. We also summarize empirical results on downstream understanding and generation benchmarks, comparing unified tokenizers with both individual task-specific models and combinations of separate tokenizers. Finally, we discuss open challenges and future directions, including the trade-off between semantic abstraction and reconstruction fidelity, scalable token vocabularies, and extensions to video and 3D domains. We further examine the evolving role of visual tokens in next-generation multimodal systems. We maintain a curated list of related works at https://github.com/Shi-qingyu/Awesome-Visual-Tokenizer.
- 23 July 2026
- Published by MDPI AG in Preprints.org
Open AccessVisual Tokenizers: A Comprehensive Survey
Visual tokenization bridges the gap between high-dimensional visual data and sequence modeling by transforming images, videos, and 3D content into compact token sequences. Recent advances in Multimodal Large Language Models (LLMs) have further underscored the critical role of visual tokenizers. Despite this rapid progress, the literature in this area remains fragmented. This work addresses this gap by presenting a comprehensive taxonomy of visual tokenizers. We review the evolution of task-specific tokenizers along two complementary dimensions: high-level tokenizers that emphasize semantic representations and low-level tokenizers that focus on pixel reconstruction and compression. Motivated by the emergence of Multimodal Large Language Models capable of supporting both understanding and generation tasks, we highlight recent advances in unified tokenizers. These models aim to integrate semantic abstraction with fine-grained visual details within a single representation. We categorize unified tokenizers into single-encoder and dual-encoder architectures and further organize them according to their semantic preservation strategies and fusion mechanisms. We also summarize empirical results on downstream understanding and generation benchmarks, comparing unified tokenizers with both individual task-specific models and combinations of separate tokenizers. Finally, we discuss open challenges and future directions, including the trade-off between semantic abstraction and reconstruction fidelity, scalable token vocabularies, and extensions to video and 3D domains. We further examine the evolving role of visual tokens in next-generation multimodal systems. We maintain a curated list of related works at https://github.com/Shi-qingyu/Awesome-Visual-Tokenizer.
- Open Access
Dense360: Dense Understanding from Omnidirectional Panoramas
Multimodal Large Language Models (MLLMs) require comprehensive visual inputs to achieve dense understanding of the physical world. While existing MLLMs demonstrate impressive world understanding capabilities through limited field-of-view (FOV) visual inputs (e.g., 70 degree), we take the first step toward dense understanding from omnidirectional panoramas. We first introduce an omnidirectional panoramas dataset featuring a comprehensive suite of reliability-scored annotations. Specifically, our dataset contains 160K panoramas with 5M dense entity-level captions, 1M unique referring expressions, and 100K entity-grounded panoramic scene descriptions. Compared to multi-view alternatives, panoramas can provide more complete, compact, and continuous scene representations through equirectangular projections (ERP). However, the use of ERP introduces two key challenges for MLLMs: i) spatial continuity along the circle of latitude, and ii) latitude-dependent variation in information density. We address these challenges through ERP-RoPE, a position encoding scheme specifically designed for panoramic ERP. In addition, we introduce Dense360-Bench, the first benchmark for evaluating MLLMs on omnidirectional captioning and grounding, establishing a comprehensive framework for advancing dense visual-language understanding in panoramic settings.
Open AccessDense360: Dense Understanding from Omnidirectional Panoramas
Multimodal Large Language Models (MLLMs) require comprehensive visual inputs to achieve dense understanding of the physical world. While existing MLLMs demonstrate impressive world understanding capabilities through limited field-of-view (FOV) visual inputs (e.g., 70 degree), we take the first step toward dense understanding from omnidirectional panoramas. We first introduce an omnidirectional panoramas dataset featuring a comprehensive suite of reliability-scored annotations. Specifically, our dataset contains 160K panoramas with 5M dense entity-level captions, 1M unique referring expressions, and 100K entity-grounded panoramic scene descriptions. Compared to multi-view alternatives, panoramas can provide more complete, compact, and continuous scene representations through equirectangular projections (ERP). However, the use of ERP introduces two key challenges for MLLMs: i) spatial continuity along the circle of latitude, and ii) latitude-dependent variation in information density. We address these challenges through ERP-RoPE, a position encoding scheme specifically designed for panoramic ERP. In addition, we introduce Dense360-Bench, the first benchmark for evaluating MLLMs on omnidirectional captioning and grounding, establishing a comprehensive framework for advancing dense visual-language understanding in panoramic settings.
- Top Cited Papers
- 19 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 47 (7) , 5918-5929
DVIS++: Improved Decoupled Framework for Universal Video Segmentation
We present the Decoupled VIdeo Segmentation (DVIS) framework, a novel approach for the challenging task of universal video segmentation, including video instance segmentation (VIS), video semantic segmentation (VSS), and video panoptic segmentation (VPS). Unlike previous methods that model video segmentation in an end-to-end manner, our approach decouples video segmentation into three cascaded sub-tasks: segmentation, tracking, and refinement. This decoupling design allows for simpler and more effective modeling of the spatio-temporal representations of objects, especially in complex scenes and long videos. Accordingly, we introduce two novel components: the referring tracker and the temporal refiner. These components track objects frame by frame and model spatio-temporal representations based on pre-aligned features. To improve the tracking capability of DVIS, we propose a denoising training strategy and introduce contrastive learning, resulting in a more robust framework named DVIS++. The proposed decoupled framework efficiently handles universal and open-vocabulary object representations, allowing DVIS++ to conduct universal and open-vocabulary video segmentation. We conduct extensive experiments on six mainstream benchmarks, including the VIS, VSS, and VPS datasets. Using a unified architecture, DVIS++ significantly outperforms state-of-the-art specialized methods on these benchmarks in closed- and open-vocabulary settings.
- 19 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 47 (7) , 5918-5929
Top Cited PapersDVIS++: Improved Decoupled Framework for Universal Video Segmentation
We present the Decoupled VIdeo Segmentation (DVIS) framework, a novel approach for the challenging task of universal video segmentation, including video instance segmentation (VIS), video semantic segmentation (VSS), and video panoptic segmentation (VPS). Unlike previous methods that model video segmentation in an end-to-end manner, our approach decouples video segmentation into three cascaded sub-tasks: segmentation, tracking, and refinement. This decoupling design allows for simpler and more effective modeling of the spatio-temporal representations of objects, especially in complex scenes and long videos. Accordingly, we introduce two novel components: the referring tracker and the temporal refiner. These components track objects frame by frame and model spatio-temporal representations based on pre-aligned features. To improve the tracking capability of DVIS, we propose a denoising training strategy and introduce contrastive learning, resulting in a more robust framework named DVIS++. The proposed decoupled framework efficiently handles universal and open-vocabulary object representations, allowing DVIS++ to conduct universal and open-vocabulary video segmentation. We conduct extensive experiments on six mainstream benchmarks, including the VIS, VSS, and VPS datasets. Using a unified architecture, DVIS++ significantly outperforms state-of-the-art specialized methods on these benchmarks in closed- and open-vocabulary settings.
- 1 November 2024
- Published by Springer Nature
- p. 446-463
Improving Video Segmentation via Dynamic Anchor Queries
- 1 November 2024
- Published by Springer Nature
- p. 446-463
Improving Video Segmentation via Dynamic Anchor Queries
- Open Access
- 7 December 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (24) , 6200
Regional Spatiotemporal Patterns of Fire in the Eurasian Subarctic Based on Satellite Imagery
The fire risks in the vast Eurasian Subarctic are increasing, raising concerns for both local and global climate systems. Although some studies have addressed this problem, their conclusions only draw from relatively lower resolution data, and the sub-regional analysis of fire patterns in this area is lacking. In this paper, using a huge amount of multi-temporal and multi-resolution remotely sensed data, derived products, and weather data between the period 2001 and 2021, we reveal several novel and recent findings concerning regional and overall fire patterns in the Eurasian Subarctic. First, we discovered that fire occurrence over the period 2001 and 2021 varied by sub-region within the Eurasian Subarctic, with perennial low fire incidence in the East European and West Siberian Plain, increasing fire incidence in the Central Siberian Plateau, and marked periodicity of fire in the East Siberian Highlands. Second, we reveal the larger scale of individual fires in the Eurasian Subarctic compared to the adjacent region to the south, with fires of longer duration (13 vs. 8 days), larger daily expansion area (7.5 vs. 3.0 km2/d), and faster propagation (442 vs. 280 m/d). Third, the northern limit of fire has extended poleward approximately 1.5° during the study period. Fourth, the start dates of fire seasons in Eurasian Subarctic, dominated by the Central Siberian Plateau, has advanced at a rate of 1.4 days per year. We also analyzed the factors resulting in the regional patterns of fire incidence including weather, human activity, land cover, and landscape structure. Our findings not only increase the knowledge of regional fire patterns and trends in Eurasian Subarctic but also will benefit the design of special fire management policies.
- 7 December 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (24) , 6200
Open AccessRegional Spatiotemporal Patterns of Fire in the Eurasian Subarctic Based on Satellite Imagery
The fire risks in the vast Eurasian Subarctic are increasing, raising concerns for both local and global climate systems. Although some studies have addressed this problem, their conclusions only draw from relatively lower resolution data, and the sub-regional analysis of fire patterns in this area is lacking. In this paper, using a huge amount of multi-temporal and multi-resolution remotely sensed data, derived products, and weather data between the period 2001 and 2021, we reveal several novel and recent findings concerning regional and overall fire patterns in the Eurasian Subarctic. First, we discovered that fire occurrence over the period 2001 and 2021 varied by sub-region within the Eurasian Subarctic, with perennial low fire incidence in the East European and West Siberian Plain, increasing fire incidence in the Central Siberian Plateau, and marked periodicity of fire in the East Siberian Highlands. Second, we reveal the larger scale of individual fires in the Eurasian Subarctic compared to the adjacent region to the south, with fires of longer duration (13 vs. 8 days), larger daily expansion area (7.5 vs. 3.0 km2/d), and faster propagation (442 vs. 280 m/d). Third, the northern limit of fire has extended poleward approximately 1.5° during the study period. Fourth, the start dates of fire seasons in Eurasian Subarctic, dominated by the Central Siberian Plateau, has advanced at a rate of 1.4 days per year. We also analyzed the factors resulting in the regional patterns of fire incidence including weather, human activity, land cover, and landscape structure. Our findings not only increase the knowledge of regional fire patterns and trends in Eurasian Subarctic but also will benefit the design of special fire management policies.