内容库 / 论文

arXiv:2607.21848 · 多模态

Closing the Loop: Training-Free Revisit Consistency for Autoregressive Generative Rendering

Closing the Loop：面向自回归生成式渲染的无训练 Revisit 一致性

Closing the Loop: Training-Free Revisit Consistency for Autoregressive Generative Rendering

类型：arxiv
标识：2607.21848
链接：https://arxiv.org/abs/2607.21848
主分类：multimodal
形态：application
TLDR：Recent conditional video generation models have shown promising potentials to transform 3D engine renderings, such as depth maps and untextured geometry, into photorealistic videos for gaming and immersive content creation. These applications require long-horizon auto-regressive generation that continuously synthesizes new frames while preserving a persistent 3D world. Auto-regressive generators synthesize video chunk by chunk with a bounded KV cache, so when the camera revisits a location after its context has been evicted, the model often regenerates inconsistent appearance, even though the
副分类：llm-infra
待LLM分类：否
标题中文：Closing the Loop：面向自回归生成式渲染的无训练 Revisit 一致性
TLDR中文：近期条件视频生成模型已展现出将 3D 引擎渲染（如深度图与无纹理几何体）转化为照片级真实视频的潜力，可应用于游戏与沉浸式内容创作。此类应用要求长时程自回归生成，在持续合成新帧的同时维持持久的 3D 世界。自回归生成器以有界 KV cache 逐 chunk 合成视频，因此当相机再次访问已从上下文中驱逐的位置时，模型常会重新生成不一致的外观，尽管该
来源文件：
/inbox/tom/_candidates/2026-07-27-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记 MIOH: Fine-Grained Multi-Image Object Hallucination Benchmark（CVPR 2026）· v2 重写覆盖原 7-08 15:50 v1

笔记 Canvas360 v2 · 几何感知预训练 × 全景图 in-context 生成的深度注入疲劳（重写）

仓库 prakash-kharvi/metaverse-screening-record

笔记 2026-07-03 轻量精读：Qwen-Image-Agent × The Verification Horizon

笔记精读与批判：SenseNova-Vision — Vision as Unified Multimodal Generation

笔记精读与批判：OPPO — Oriented Pickup Preference Optimization for Multimodal Hallucination Mitigation

arXiv

2607.21848

TLDR

近期条件视频生成模型已展现出将 3D 引擎渲染（如深度图与无纹理几何体）转化为照片级真实视频的潜力，可应用于游戏与沉浸式内容创作。此类应用要求长时程自回归生成，在持续合成新帧的同时维持持久的 3D 世界。自回归生成器以有界 KV cache 逐 chunk 合成视频，因此当相机再次访问已从上下文中驱逐的位置时，模型常会重新生成不一致的外观，尽管该Recent conditional video generation models have shown promising potentials to transform 3D engine renderings, such as depth maps and untextured geometry, into photorealistic videos for gaming and immersive content creation. These applications require long-horizon auto-regressive generation that continuously synthesizes new frames while preserving a persistent 3D world. Auto-regressive generators synthesize video chunk by chunk with a bounded KV cache, so when the camera revisits a location after its context has been evicted, the model often regenerates inconsistent appearance, even though the

相关解读

arXiv ↗ 中文全文