| 图像匹配与稀疏对应 |
FlexGlue: Sparse-to-Dense Feature Matching and Visual Disambiguation |
ETH Zurich，2026-10-23 |
未复现（高成本资料） |
| 统一动作表示与条件注意力 |
UniAction: Unified Action Conditioning for World Models |
Stanford University，2026-10-23 |
uniaction |
| 代码、视频与动作联合生成 |
Agentic World Action Models |
UC Berkeley，2026-10-21 |
未复现（高成本资料） |
| 单次前向结构化重排 |
Multimodal Reranking Is a System One Decision |
Aalto University，2026-10-20 |
system-one-rerank |
| 视觉证据约束蒸馏 |
CV-OPD: Contrastive Vision On-policy Distillation |
Xiamen University，2026-10-09 |
cv-opd |
| 视觉 token 压缩与动态证据恢复 |
Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning |
City University of Hong Kong，2026-10-08 |
vimod |
| 冻结中间层探针与验证器自举 |
Unsupervised Multimodal Reward Modeling |
Massachusetts Institute of Technology，2026-10-08 |
umrm |
| 视觉 token 压缩与动态证据恢复 |
V-CoLA: Vision Token Compression with Linear Attention |
Alibaba Cloud Computing，2026-10-08 |
vcola |
| 视觉状态转移监督 |
WOVEN: Weaving Visual World Modeling into Multimodal LLMs |
Northwestern University，2026-10-08 |
woven |
| 视觉 token 与跨模态检索 |
EmbeddingGemma 2: Unified and Efficient Multimodal Representations |
Google，2026-10-06 |
embedding-gemma2 |
| 视觉 token 与跨模态检索 |
StackTok: Accelerating VLMs Inference with Budget-Adaptive Visual Token Selection |
Sichuan University，2026-09-15 |
stacktok |
| 视频理解与 token 压缩 |
VideoMM: Adaptive Macro-Micro Inference for Efficient Video MLLMs |
University of Science and Technology of China，2026-09-15 |
videomm |
| 音频描述与对比排序 |
From Visual Cues to Spoken Narration: Rethinking Audio Description |
TU Darmstadt & hessian.AI，2026-09-01 |
cue2narrate-drl |
| 生成辅助监督与理解增强 |
Generation as Auxiliary Supervision: Enhancing Visual Understanding at Zero Inference Overhead via Decoupled Embedding Prediction |
ByteDance，2026-08-12 |
gas |
| 多实例编辑与注意力隔离 |
Editing Everything Everywhere All at Once |
University of Modena and Reggio Emilia，2026-06-30 |
mice |
| 视频嵌入统一评测 |
MVEB: Massive Video Embedding Benchmark |
Harvard University，2026-06-12 |
未复现（高成本资料） |
| 低资源语音翻译评测 |
NaijaS2ST: A Multi-Accent Benchmark for Speech-to-Speech Translation in Low-Resource Nigerian Languages |
Mila - Quebec AI Institute，2026-04-17 |
未复现（高成本资料） |
| 高效视觉 token 压缩 |
SmolVLM: Redefining small and efficient multimodal models |
Hugging Face，2025-04-07 |
smolvlm |
| 对比预训练与自蒸馏 |
SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features |
Google DeepMind，2025-02-20 |
siglip2 |
| 视觉 token 与跨模态检索 |
Visual Instruction Tuning |
University of Wisconsin-Madison，2023-04-17 |
llava |
| 跨模态连接器与冻结骨干 |
BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models |
Salesforce Research，2023-01-30 |
blip2 |
| 视觉 token 与跨模态检索 |
Learning Transferable Visual Models From Natural Language Supervision |
OpenAI，2021-02-26 |
clip |