Computer Vision Research Map
A computer-vision research wiki for newcomers
Generative vision, video understanding and sports AI: paper notes, topic reviews and concrete research entry points. The map below connects them — pick a node to start.
Research relation network
48 knowledge nodes · 161 explicit relations
Wheel or pinch to zoom · drag with one finger · select a node for details
Accessible node index
Use this list to open every map node directly.
- Computer Vision OverviewFoundations
- Datasets, Metrics, and Benchmark ReliabilityFoundations
- Diffusion ModelsFoundations
- Foundations of Core Vision TasksFoundations
- Foundations of Visual Generative ModelsFoundations
- Video Representation and Temporal ModelingFoundations
- Vision-Language and Multimodal Representation FoundationsFoundations
- Visual Representations and BackbonesFoundations
- 3D Gaussian Splatting: Representation, Optimization, Geometry, and Dynamic ScenesResearch
- Diffusion Efficiency EngineeringResearch
- Generative Model EvaluationResearch
- Image EditingResearch
- Image GenerationResearch
- Image-Text RetrievalResearch
- Sports AI Video UnderstandingResearch
- Video EditingResearch
- Video GenerationResearch
- Video UnderstandingResearch
- Vision-LanguageResearch
- Sports AI Research RoadmapRoadmaps
- CLIP 系列模型（CLIP Family）Entities
- ColPali 与视觉文档检索（ColPali）Entities
- DINO 家族：自监督视觉表示与预训练骨干Entities
- M-BEIR 多模态检索评测（M-BEIR）Entities
- SportsMOTEntities
- 图像生成与编辑统一建模Entities
- 图文检索与组合性评测集（Image-Text Retrieval Benchmarks）Entities
- 扩散 TransformerEntities
- 视频编辑理解Entities
- DiT and Transformer-Style Denoisers Are Now Major Scalable Generative BackbonesClaims
- General Vid-LLMs Still Struggle With Editing-Element Recognition, Reasoning, and Functional JudgmentClaims
- Unified Image Models Show Multi-Task and Workflow Benefits; Overall Performance Leadership Remains UnverifiedClaims
- 双塔负责可扩展召回，细粒度模块负责精排与关系匹配Claims
- 图文检索的主要瓶颈仍包括数据质量、分布与概念覆盖Claims
- 标准图文检索分数不足以证明组合性理解Claims
- 组合式查询是图文检索中可落地且仍有缺口的研究切口Claims
- 统一多模态向量需要同时验证任务意图、跨域泛化与检索成本Claims
- Can Motion-Defined Text Become a Reliable Human–Machine Perception Benchmark?Questions
- DINO 密集特征如何迁移到三维高斯语义分割？Questions
- 图像编辑的进步主要来自数据扩展还是架构升级Questions
- 图文检索哪个子方向最适合中等算力的研究生项目？Questions
- 如何审计图文检索中的预训练重叠、概念频率与 benchmark 污染？Questions
- 怎样构造既难又不错误的图文检索负例？Questions
- 现有评测是否真的刻画了视频编辑理解能力Questions
- 细粒度图文对齐如何兼顾索引效率？Questions
- 统一图像模型与专用或模块化流水线如何分工Questions
- 统一多模态向量是否真的能跨任务、跨语言泛化？Questions
- 羽毛球挥拍动作纠正 demo 应如何定义Questions