Tag
AcrossVAM2.0 提出一种冲突感知的点对应监督方法，用于参数高效的视频扩散模型适配，通过置信度加权跟踪头、零初始化残差反馈和非对称 PCGrad，在保持预训练外观先验的同时提升对应一致性，但并不均匀改善视频质量。
FreeFlow is a hierarchical transformer for optical flow estimation that eliminates task-specific inductive biases and achieves state-of-the-art accuracy on benchmarks like Sintel and KITTI-2015.
This paper proposes distilling deep optical flow stereo methods into a single-satellite model for efficient retrieval of dense three-dimensional wind fields, improving accuracy over traditional atmospheric motion vectors in certain spectral bands.
RECAP-Forcing improves long video generation by indexing memory based on appearance novelty instead of recency, preserving key-value caches for new content to maintain consistency and quality without additional training.
This article discusses the technical challenges of quality control (QC) in synthetic videos, including geometric consistency, temporal consistency, and multimodal cross-validation, which make synthetic videos prone to visible flaws.
An open-source web application using computer vision to automatically pilot a Steam Controller onto its magnetic charging puck via optical flow and WebHID.