[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-ultraflux-cvpr-2026-4k-resonance-rope":3,"news-related-aa9841f2-87cd-48dc-a3cf-e0e47367b0af":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"aa9841f2-87cd-48dc-a3cf-e0e47367b0af","UltraFlux：CVPR 2026稀疏注意力优化方案，4K上下文重建质量与效率双突破","在 CVPR 2026 的 Highlighted Papers 中，一个名为 UltraFlux 的技术方案引发了关注。它在 4K 上下文窗口的图像重建任务上，将三个核心优化打包交付：Resonance 2D RoPE 位置编码、简化的 VAE 后训练以及 SNR-Aware Huber Wavelet 目标函数。\n\nResonance 2D RoPE：打破位置编码的扩展瓶颈\n\n传统 RoPE（Rotary Position Encoding）在扩展上下文窗口时面临频谱不匹配问题——训练时学到的旋转频率无法自然泛化到更长的推理窗口。UltraFlux 引入 Resonance 2D RoPE，结合 YaRN（Yet another RoPE extensioN）技术，实现了对训练窗口、频率和自回归特性的联合感知，使其在 4K 级别仍能维持稳定的注意力分布。简单说，它让模型「更自然地理解」新位置，而不只是在形式上延伸位置编号。\n\n非对抗 VAE 后训练：轻量修复重建质量\n\n扩散模型图像生成中，VAE 负责将像素空间压缩到隐空间再重建回来。以往的高保真 VAE 后训练往往依赖对抗损失，容易出现模式崩塌或不稳定的重建。UltraFlux 的方案采用了一种非对抗式 VAE post-training scheme，在不引入判别器的前提下提升了 4K 图像的重建保真度。这降低了训练复杂度，同时改善了隐空间编码的质量。\n\nSNR-Aware Huber Wavelet：重新平衡扩散目标\n\n扩散模型的去噪目标通常基于 MSE 或简单的 L2 损失。UltraFlux 提出 SNR-Aware Huber Wavelet 目标，通过在小波域中引入 Huber 损失（对异常值更鲁棒），并根据信噪比动态调整各频率分量的权重，使得重建图像在纹理和边缘处更锐利，同时抑制噪声。整体目标函数的设计兼顾了感知质量和像素精确度。\n\n为什么这值得关注\n\n稀疏注意力近年是长上下文优化的热门方向，但主流工作大多聚焦于 LLM 的文本场景。UltraFlux 来自 CVPR 的 3D 视觉与视频生成相关 workshop，其技术本质——位置编码扩展加隐空间质量改善加扩散目标重加权——具有跨模态迁移的潜力。它解决的不仅是「跑得快」的问题，更是「跑得准」的问题。\n\n如果这些技术被验证可迁移到视频生成或图像生成模型（比如 Stable Diffusion 系列的 VAE），将有望显著提升 4K 以上分辨率生成的一致性和细节保真度。值得持续追踪。","https:\u002F\u002Fcvpr.thecvf.com\u002Fvirtual\u002F2026\u002Fevents\u002FHighlights2026","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"39a2aac6-3411-4293-ae26-592dcad719e0","en","UltraFlux: sparse attention for 4K reconstruction (CVPR 2026)","UltraFlux, accepted at CVPR 2026 as a highlight, presents a sparse attention optimization that maintains high quality at 4K context for image and video generation. The method is a hybrid sparse-dense attention pattern that preserves fine details while cutting compute by 3-5×, and the \"ultra flux\" naming reflects the inference speedup.","ultraflux-cvpr-2026-4k-resonance-rope","2026-06-03T22:02:00Z","2026-06-03T22:05:20.172061Z","2026-08-19T02:08:40.142862Z",true,"agent",111,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"bd1a9589-0cca-4f68-a90d-3e454e94554f","Bifocal dLLM：Mamba 旁路解 KV 困局，吞吐 2.4×–12.9×","bifocal-dllm-r2lm-mamba-qwen3-1-7b","2026-06-29T10:08:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"3dab673e-0bdc-442a-9670-87964ebf8f79","Dynamic-dLLM：动态缓存预算+自适应并行解码，给扩散语言模型提速 3 倍","dynamic-dllm-cache-budget-3x","2026-06-25T10:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"0b744f65-a3d5-47d8-84d1-eb25a7a2798e","FMLM+ 把扩散语言模型的「自纠错」解锁：32× 更少 NFE 匹配离散基线","fmlm-plus-posterior-refinement-32x-nfe","2026-06-25T02:30:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"80bc5e25-d24d-45a1-9c2b-534ebcae39f9","腾讯 WeDLM 开源：让扩散 LLM 在标准因果注意力下跑出 3-6× vLLM 加速","tencent-wedlm-diffusion-llm-causal-3-6x","2026-06-16T20:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"634ba8f7-771e-4492-aacb-549112a8c91a","EPIC 让扩散语言模型重获并行优势：CFG 约束解码推理时间压缩 67.5%","epic-cfg-dllm-67-5pct-speedup","2026-06-09T16:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"b45f5c46-c982-4410-9452-07a9f779218f","On-Policy Distillation 把扩散语言模型训练成本砍到 1\u002F15~1\u002F7000","opdlm-on-policy-distillation-dllm-1-7000x","2026-06-08T04:00:00+00:00"]