![](https://img-blog.csdnimg.cn/20201014180756926.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
行为识别
文章平均质量分 94
h137437
这个作者很懒,什么都没留下…
展开
-
MoViNets: Mobile Video Networks for Efficient Video Recognition
MoViNetsAbstract1 Introduction2. Related Work3. Mobile Video Networks (MoViNets)3.1. Searching for MoViNet3.2. The Stream Buffer with Causal Operations3.2.1 Causal Operations3.2.2 Training and Inference with Stream Buffers3.3. Temporal Ensembles4. Experime翻译 2021-04-19 10:12:57 · 3003 阅读 · 0 评论 -
Revisiting ResNets: Improved Training and Scaling Strategies
ResNet-RSAbstract1 Introduction2. Characterizing Improvements on ImageNet4. MethodologyExperimentsConclusion备注: 如有侵权,立即删除code: https://github.com/tensorflow/tpu/tree/master/models/official/resnet/resnet_rssource: 2021Abstract新的计算机视觉架构占据了焦点,但模型架构的影响往往与翻译 2021-04-13 15:09:27 · 789 阅读 · 0 评论 -
Is Space-Time Attention All You Need for Video Understanding?
TimeSformerAbstract1 Introduction3 The TimeSformer ModelExperiments4.1. Analysis of Self-Attention Schemes4.2. Varying the Number of Tokens in Space and Time4.3. The Importance of Pretraining and Dataset Scale4.4. Comparison to the State-of-the-Art4.5. Lon翻译 2021-04-06 16:58:30 · 829 阅读 · 0 评论 -
Mutual Modality Learning for Video Action Classification
MMLAbstract1 Introduction3 Proposed MethodExperimentsConclusionhttps://arxiv.org/pdf/2011.02543v1.pdf备注: 如有侵权,立即删除code: https://github.com/papermsucode/mutual-modality-learningsource: 2020Abstract1 Introduction3 Proposed MethodExperimentsConclusio翻译 2020-12-13 20:14:13 · 266 阅读 · 0 评论 -
MotionSqueeze: Neural Motion Feature Learning for Video Understanding
MotionSqueezeAbstract1 Introduction3 Proposed MethodExperimentsConclusionhttps://arxiv.org/pdf/2011.02543v1.pdf备注: 如有侵权,立即删除code: https://github.com/arunos728/MotionSqueezesource: ECCV2020Abstract运动在视频理解领域非常重要,大多数视频分类的神经网络模型通过现有的光流提取方法来利用运动信息。因为计算帧与帧翻译 2020-12-11 14:25:04 · 1626 阅读 · 0 评论 -
TSM: Temporal Shift Module for Efficient Video Understanding
TSMAbstract1 Introduction3 Proposed MethodExperimentsConclusionhttps://arxiv.org/abs/1811.08383备注: 如有侵权,立即删除code: https://github.com/mit-han-lab/temporal-shift-modulesource: ICCV2019Abstract视频流的爆炸性增长给高准确率和低计算成本的视频理解的带来了很大的挑战。传统的2D CNNs计算量很小但是无法捕获时序关系翻译 2020-12-09 15:29:53 · 627 阅读 · 0 评论 -
PAN: Towards Fast Action Recognition via Learning Persistence of Appearance
PANAbstract1 Introduction3 Proposed MethodExperimentsConclusion备注: 如有侵权,立即删除code: https://github.com/zhang-can/PAN-PyTorchsource: 2020Abstract高效建模视频中的动态运动信息对于行为识别任务非常重要。大部分表现好的方法都依赖于用密集光流代表行为特征。尽管结合光流和RGB帧能够取得更好的效果,光流提取是非常耗时的。这无疑是不利于实时行为识别的。在本文中,我们解除了翻译 2020-12-08 15:08:56 · 847 阅读 · 1 评论 -
Late Temporal Modeling in 3D CNN Architectures with BERT for Action Recognition
R2+1D-BERTAbstract1 Introduction3 Proposed Method3.1 BERT-based Temporal Modeling with 3D CNNs for Action Recognition3.2 Proposed Feature Reduction Blocks: FRAB & FRMB3.3 Proposed BERT Implementations on SlowFast ArchitectureExperimentsConclusion备注: 机翻译 2020-11-29 22:59:34 · 1310 阅读 · 0 评论