Haonian Ji

Agent0-VL is a self-evolving vision-language agent that integrates tool use into reasoning, self-evaluation, and self-repair. Without human annotations or external reward models, it continually improves its evidence-grounded multimodal reasoning and achieves strong results on geometric problem solving and visual scientific analysis.

Recent advancements in video generation have significantly improved the ability to synthesize videos from text instructions. However, existing models still struggle with key challenges such as instruction misalignment, content hallucination, safety concerns, and bias. Addressing these limitations, we introduce MJ-BENCH-VIDEO, a large-scale video preference benchmark designed to evaluate video generation across five critical aspects: Alignment, Safety, Fineness, Coherence & Consistency, and Bias & Fairness. This benchmark incorporates 28 fine-grained criteria to provide a comprehensive evaluation of video preference. Building upon this dataset, we propose MJ-VIDEO, a Mixture-of-Experts (MoE)-based video reward model designed to deliver fine-grained reward. MJ-VIDEO can dynamically select relevant experts to accurately judge the preference based on the input text-video pair. This architecture enables more precise and adaptable preference judgments. Through extensive benchmarking on MJ-BENCH-VIDEO, we analyze the limitations of existing video reward models and demonstrate the superior performance of MJ-VIDEO in video preference assessment, achieving 17.58% and 15.87% improvements in overall and fine-grained preference judgments, respectively. Additionally, introducing MJ-VIDEO for preference tuning in video generation enhances the alignment performance.

Existing video benchmarks often resemble image-based benchmarks, with questions like “What color is the woman’s dress in this video?” or “What’s the weather like in the video?” where models can answer using just a single frame or text, limiting the assessment of large vision language models (LVLMs) in terms of video perception and temporal reasoning abilities. To address this, we introduce GLIMPSE, a benchmark specifically designed to evaluate whether LVLMs truly understand video content. GLIMPSE includes 3,269 videos and over 4,342 highly visual-centric questions across 11 categories, such as Trajectory, Temporal QA, Forensics Detection, and Video Sorting, all meticulously curated and crafted by human annotators. Many of these questions require watching the entire video and reasoning based on video context to answer accurately, as they cannot be addressed using only a few frames or text information. Human testing on GLIMPSE achieves an average accuracy of 94.82%, but GLIMPSE poses significant challenges for existing LVLMs. Even the best-performing model, Gemini 1.5 Pro, reaches only 56.98% accuracy, indicating that current LVLMs still have a long way to go in truly understanding and interpreting video content.

In solving multimodal multiobjective optimization problems (MMOPs), changes in objective and decision space need to be considered. However, many multimodal multiobjective evolutionary algorithms (MMOEAs) tend to prioritize diversity in the objective space during the optimization process, often leading to suboptimal diversity and distribution in the decision space. This study proposes the Dynamic-Niche-Based Two-Stage Evolution (DNTE) approach to address this issue. DNTE employs differential evolutionary strategies in two stages, utilizing dynamic niche to enhance solution viability and optimize decision space distribution at different phases. In the first stage, dynamic niche is used to restrict the dominance range of Pareto domination. In contrast, in the second stage, it is integrated with differential evolution to refine the distribution of the decision space. The combination of DNTE with DN-NSGA-II enhances the capability of solving MMOPs and demonstrates superior performance compared to other evolutionary algorithms on MMF and MMMOP benchmark functions