GitHub is where people build software. More than 150 million people use GitHub to discover, fork, and contribute to over 420 million projects.
Explore Why GitHub All features Documentation GitHub Skills Blog Solutions By company size Enterprises Small and medium teams Startups Nonprofits By use case DevSecOps DevOps CI/CD View all use cases By industry Healthcare Financial services Manufacturing Government View all industries ...
代码(刚刚开源): https://github.com/OpenGVLab/UniFormerV2 论文下载链接:https://arxiv.org/abs/2211.09552 简单介绍一下我们最近放出来的工作UniFormerV2,方法受UniFormer的启发,设计了通用高效的时序建模模块,可以无缝插入到各种开源预训练图像ViT中,显著增强模型对时序信息的处理能力。为进一步提升主流benchmark上的...
Code: https://github.com/OpenGVLab/UniFormerV2 简单介绍一下我们最近放出来的工作UniFormerV2,方法受UniFormer的启发,设计了通用高效的时序建模模块,可以无缝插入到各种开源预训练图像ViT中,显著增强模型对时序信息的处理能力。为进一步提升主流benchmark上的性能,我们将K400/K600/K700进行数据清洗,得到更精简的K710数...
Ranked #1 onAction Classification on ActivityNet(using extra training data) Get a GitHub badge TaskDatasetModelMetric NameMetric ValueGlobal RankUses Extra Training DataBenchmark Action ClassificationActivityNetUniFormerV2-LTop 1 Accuracy94.7# 1 Compare ...
https://github.com/OpenGVLab/UniFormerV2 Motivation ▲ motivation 去年在做UniFormer时,我们时常觉得实验的周期太长了。由于 UniFormer 是全新的框架,每次调整结构,我们都需要先经过一轮 ImageNet 图像预训练,之后将模型展开后,在 Kinetics 等视频数据集进行二次微调。根据结果反馈,再进行结构的改进。
代码地址:https://github.com/OpenGVLab/Un 论文链接:UniFormerV2: Spatiotemporal Learning by Arming Image ViTs with Video UniFormer 更多资料:UniFormerV2开源,K400首次90%准确率 | 基于ViT的高效视频识别,8数据集SOTA 如果你觉得我们分享的内容还不错,请不要吝啬给我们一些鼓励:点赞、喜欢或者分享给你的小伙伴...
AddRemoveMark official No code implementations yet. Submityour code now Video Understanding Datasets Edit Submitresults from this paperto get state-of-the-art GitHub badges and help the community compare results to other papers. Methods Edit
论文链接: https://arxiv.org/abs/2211.09552 代码链接: https://github.com/OpenGVLab/UniFormerV2 基于CLIP提供的视觉编码器,我们的UniFormerV2最终在8个主流benchmark都取得了SOTA结果,包括场景相关数据集(短时的K400/K600/K700和Moments in Time,以及长时的ActivityNet和HACS),和时序相关数据集(Something-Somethi...
https://github.com/OpenGVLab/UniFormerV2 Motivation ▲ motivation 去年在做UniFormer时,我们时常觉得实验的周期太长了。由于 UniFormer 是全新的框架,每次调整结构,我们都需要先经过一轮 ImageNet 图像预训练,之后将模型展开后,在 Kinetics 等视频数据集进行二次微调。根据结果反馈,再进行结构的改进。