Action Reconition CVPR 2017

Action Recognition in CVPR 2017

CVPR2017的paper list

本文總結了CVPR2017中關於行為識別的一些文章，action detection的文章並未總結，附上pdf鏈接~

很多文章代碼都沒有放出來，找到了會更新的哦~

Analyzing Humans in Images

• Transition Forests: Learning Discriminative Temporal Transitions for Action Recognition and Detection
• Scene Flow to Action Map: A New Representation for RGB-D Based Action Recognition With Convolutional Neural Networks
• Spatio-Temporal Naive-Bayes Nearest-Neighbor (ST-NBNN) for Skeleton-Based Action Recognition
• 該文章沒有上deep learning，作者的理由是，我處理的不是完整的圖片而是骨架，骨架信息並不像圖片，有成千上萬個像素，只有上十個關節信息，因此不需要端到端的複雜模型，非參數模型也能搞定
• Procedural Generation of Videos to Train Deep Action Recognition Networks
• Learning and Refining of Privileged Information-Based RNNs for Action Recognition From Depth Sequences
• Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset（強推，HMDB51上的準確率一下子提升到了80%多，上了64塊GPU，DeepMind做出來的，感覺任何事情谷歌插一腳，大家就沒法玩了orz）
• 文中有個圖比較直觀的反應了幾種網路的區別，其中K表示一個視頻中的總幀數，N代表幾個相鄰幀

• Asynchronous Temporal Fields for Action Recognition
• AdaScan: Adaptive Scan Pooling in Deep Convolutional Neural Networks for Human Action Recognition in Videos
• Modeling Temporal Dynamics and Spatial Configurations of Actions Using Two-Stream Recurrent Neural Networks
• Deep Sequential Context Networks for Action Prediction

Video Analytics

• Zero-Shot Action Recognition With Error-Correcting Output Codes
• Spatiotemporal Pyramid Network for Video Action Recognition
• Spatio-Temporal Vector of Locally Max Pooled Features for Action Recognition in Videos
• UntrimmedNets for Weakly Supervised Action Recognition and Detection
• Spatiotemporal Multiplier Networks for Video Action Recognition
• ActionVLAD: Learning Spatio-Temporal Aggregation for Action Classification

Spotlight 1-2B Analyzing Humans 1

• Deep Learning on Lie Groups for Skeleton-Based Action Recognition
• 提出了一個新的神經網路LieNet，學習基於李群的3D骨架特徵來進行動作識別
• 1.將李群結構和神經網路結合起來，相比於傳統的神經網路結構，網路結構為了適應李群做了相應的調整，添加了RotMap Layer、RotPooling Layer、. LogMap Layer
• 2.在這個結構中，為了使用反向傳播演算法，隨機梯度下降法也做了相應探索（李群比較難....沒有細讀）

3D Computer Vision

• Global Context-Aware Attention LSTM Networks for 3D Action Recognition

Applications

• Modeling Sub-Event Dynamics in First-Person Action Recognition

others

• Multi-Task Clustering of Human Actions by Sharing Information
• Binary Coding for Partial Action Analysis With Limited Observation Ratios

推薦閱讀：