Despite advancements in Video Large Language Models (Vid-LLMs) improving multimodal understanding, challenges persist in streaming video reasoning due to its reliance on contextual information. Existing paradigms feed all available historical contextual information into Vid-LLMs, resulting in a significant computational burden for visual data processing. Furthermore, the inclusion of irrelevant context distracts models from key details. This paper introduces a challenging task called Context-guided Streaming Video Reasoning (CogStream), which simulates real-world streaming video scenarios, requiring models to identify the most relevant historical contextual information to deduce answers for questions about the current stream. To support CogStream, we present a densely annotated dataset featuring extensive and hierarchical question-answer pairs, generated by a semi-automatic pipeline. Additionally, we present CogReasoner as a baseline model. It effectively tackles this task by leveraging visual stream compression and historical dialogue retrieval. Extensive experiments prove the effectiveness of this method.

Kangyu Wang, Zhiyun Jiang, Haibo Feng, Weijia Zhao, Lin Liu, Jianguo Li, Zhenzhong Lan, Weiyao Lin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

In extended reality (XR) applications, enhancing user perception often necessitates head-mounted display (HMD) removal. However, existing methods suffer from low time performance and suboptimal reconstruction quality. In this paper, we propose a half face mapping 3D Gaussian splatting avatar based HMD removal method (HFM-GS), which can perform real-time and high-fidelity online restoration of the complete face in HMD-occluded videos for XR applications after a short un-occluded face registration. We establish a mapping field between the upper and lower face Gaussians to enhance the adaptability to deformation. Then, we introduce correlation weight-based sampling to improve time performance and handle variations in the number of Gaussians. At last, we ensure model robustness through Gaussian Segregation Strategy. Compared to two state-of-the-art methods, our method achieves better quality and time performance. The results of the user study show that fidelity is significantly improved with our method.

We propose a new scene-aware foveated rendering method, which incorporates the scene awareness and characteristics of the human visual system into the mapping-based foveated rendering framework. First, we generate the conservative visual importance map that encodes the visual features of the scene, visual acuity, and gaze motion. Second, we construct the pixel size control map using a convolution kernel method. Third, we utilize the pixel size control map to guide the foveated rendering. At last, a temporal coherent refinement strategy is used to maintain the smooth foveated rendering for the adjacent frames. Compared to the state-of-the-art mapping-based foveated rendering methods using the same compression ratio, our method achieves smaller MSE, higher PSNR, and SSIM in the fovea, periphery, salient regions, and the whole image. We also conducted user studies, and the results proved that the perceptual quality of our method has a high visual similarity with the around truth rendered with the full resolution.

Global ocean data assimilation is a crucial technique to estimate the actual oceanic state by combining numerical model outcomes and observation data, which is widely used in climate research. Due to the imbalanced distribution of observation data in global ocean, the parallel efficiency of recent methods suffers from workload imbalance. When massive GPUs are applied for global ocean data assimilation, the workload imbalance becomes more severe, resulting in poor scalability. In this work, we propose a novel adaptive workload-balance scheduling strategy, Bassimilation, which successfully estimates the total workload prior to execution and ensures a balanced workload assignment. Further, we design a parallel dynamic programming approach to accelerate the schedule decision, and develop a factored dataflow to exploit the parallel potential of GPUs. Evaluation demonstrates that our algorithm outperforms the state-of-the-art method by up to 9.1× speedup. This work is the first to scale global ocean data assimilation to 4, 000 GPUs.

We propose an eye-shaped keyboard for high-speed text entry in virtual reality (VR), having the shape of dual eyes with characters arranged along the curved eyelids, which ensures low density and short spacing of the keys. The eye-shaped keyboard references the QWERTY key sequence, allowing the users to benefit from their experience using the QWERTY keyboard. The user interacts with an eye-shaped keyboard using rays controlled with both the hands. A character can be entered in one step by moving the rays from the inner eye regions to regions of the characters. A high-speed auto-complete system was designed for the eye-shaped keyboard. We conducted a pilot study to determine the optimal parameters, and a user study to compare our eye-shaped keyboard with the QWERTY and circular keyboards. For beginners, the eye-shaped keyboard performed significantly more efficiently and accurately with less task load and hand movement than the circular keyboard. Compared with the QWERTY keyboard, the eye-shaped keyboard is more accurate and significantly reduces hand translation while maintaining similar efficiency. Finally, to evaluate the potential of eye-shaped keyboards, we conducted another user study. In this study, the participants were asked to type continuously for three days using the proposed eye-shaped keyboard, with two sessions per day. In each session, participants were asked to type for 20min, and then their typing performance was tested. The eye-shaped keyboard was proven to be efficient and promising, with an average speed of 19.89 words per minute (WPM) and mean uncorrected error rate of 1.939%. The maximum speed reached 24.97 WPM after six sessions and continued to increase.