菜单
Build vision-language agents that seamlessly integrate visual reasoning with dynamic tool manipulation (search, cropping) through reinforcement learning, achieving state-of-the-art performance on fine-grained visual understanding tasks—surpassing proprietary models like Gemini and GPT.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ADu2021/skillXiv --skill sensenovamars-multimodal-reasoning命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | sensenovamars-multimodal-reasoning |
| title | SenseNova-MARS: Empowering Multimodal Agentic Reasoning and Search via Reinforcement Learning |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2512.24330 |
| keywords | ["Multimodal Agents","Vision-Language Models","Agentic Reasoning","Tool Use","RL Training"] |
| description | Build vision-language agents that seamlessly integrate visual reasoning with dynamic tool manipulation (search, cropping) through reinforcement learning, achieving state-of-the-art performance on fine-grained visual understanding tasks—surpassing proprietary models like Gemini and GPT. |
SenseNova-MARS is a multimodal agentic framework enabling Vision-Language Models (VLMs) to perform sophisticated visual reasoning tasks through interleaved tool manipulation and continuous reasoning.
Core Innovation: While VLMs excel at text-oriented reasoning and isolated tool calls, they struggle with coordinated tool sequences on knowledge-intensive visual tasks. SenseNova-MARS enables human-like proficiency in dynamically invoking multiple tools (search, image cropping, text search) to solve complex visual understanding problems.
Three complementary tools enable sophisticated visual problem-solving:
Image Search Tool
Text Search Tool
Image Crop Tool
The agent operates through repeated cycles of:
To enable reliable tool invocation and reasoning, SenseNova-MARS introduces BN-GSPO algorithm:
Key Components:
Training Stability Features:
HR-MMSearch (proprietary high-resolution benchmark):
MMSearch Benchmark:
Tool Invocation in Agent Loop:
def solve_visual_task(image: PIL.Image, question: str) -> str:
"""Solve visual understanding task through tool orchestration."""
# Initial analysis
context = vlm_analyze(image, question)
# Agentic loop: iteratively invoke tools
reasoning_steps = []
for _ in range(max_steps):
# Determine next action
action = vlm_decide_next_action(context, reasoning_steps)
if action == "image_search":
ref_images = search_visual_db(context.search_query)
context.add_reference_images(ref_images)
elif action == "crop_region":
roi = vlm_identify_roi(image)
cropped = image.crop(roi)
detail = vlm_analyze_detail(cropped)
context.add_detail(detail)
elif action == "text_search":
facts = search_knowledge_base(context.search_query)
context.add_knowledge(facts)
elif action == "answer":
return vlm_synthesize_answer(context, image)
reasoning_steps.append(action)
return vlm_synthesize_answer(context, image)
vs. Image-Only Analysis:
vs. Text-Only Approaches:
vs. Isolated Tool Invocation:
Use SenseNova-MARS when:
When NOT to use:
HR-MMSearch Dataset:
All code, models, and datasets will be released for community research.