We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets, 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets, and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8% accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.

/pdf/learning-spatiotemporal-features-with-3d-convolutional-4poxybc3zp.pdf

Learning Spatiotemporal Features with 3D Convolutional Networks

We investigate architectures of discriminatively trained deep Convolutional Networks (ConvNets) for action recognition in video. The challenge is to capture the complementary information on appearance from still frames and motion between frames. We also aim to generalise the best performing hand-crafted features within a data-driven learning framework.

Our contribution is three-fold. First, we propose a two-stream ConvNet architecture which incorporates spatial and temporal networks. Second, we demonstrate that a ConvNet trained on multi-frame dense optical flow is able to achieve very good performance in spite of limited training data. Finally, we show that multitask learning, applied to two different action classification datasets, can be used to increase the amount of training data and improve the performance on both. Our architecture is trained and evaluated on the standard video actions benchmarks of UCF-101 and HMDB-51, where it is competitive with the state of the art. It also exceeds by a large margin previous attempts to use deep nets for video classification.

https://proceedings.neurips.cc/paper/2014/file/00ec53c4682d36f5c4359f4ae7bd7ba1-Paper.pdf

Two-Stream Convolutional Networks for Action Recognition in Videos

Large-scale Video Classiﬁcation with Convolutional Neural Networks

The paucity of videos in current action classification datasets (UCF-101 and HMDB-51) has made it difficult to identify good video architectures, as most methods obtain similar performance on existing small-scale benchmarks. This paper re-evaluates state-of-the-art architectures in light of the new Kinetics Human Action Video dataset. Kinetics has two orders of magnitude more data, with 400 human action classes and over 400 clips per class, and is collected from realistic, challenging YouTube videos. We provide an analysis on how current architectures fare on the task of action classification on this dataset and how much performance improves on the smaller benchmark datasets after pre-training on Kinetics. We also introduce a new Two-Stream Inflated 3D ConvNet (I3D) that is based on 2D ConvNet inflation: filters and pooling kernels of very deep image classification ConvNets are expanded into 3D, making it possible to learn seamless spatio-temporal feature extractors from video while leveraging successful ImageNet architecture designs and even their parameters. We show that, after pre-training on Kinetics, I3D models considerably improve upon the state-of-the-art in action classification, reaching 80.2% on HMDB-51 and 97.9% on UCF-101.

/pdf/quo-vadis-action-recognition-a-new-model-and-the-kinetics-1jacvbzo22.pdf

Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset

Feature trajectories have shown to be efficient for representing videos. Typically, they are extracted using the KLT tracker or matching SIFT descriptors between frames. However, the quality as well as quantity of these trajectories is often not sufficient. Inspired by the recent success of dense sampling in image classification, we propose an approach to describe videos by dense trajectories. We sample dense points from each frame and track them based on displacement information from a dense optical flow field. Given a state-of-the-art optical flow algorithm, our trajectories are robust to fast irregular motions as well as shot boundaries. Additionally, dense trajectories cover the motion information in videos well. We, also, investigate how to design descriptors to encode the trajectory information. We introduce a novel descriptor based on motion boundary histograms, which is robust to camera motion. This descriptor consistently outperforms other state-of-the-art descriptors, in particular in uncontrolled realistic videos. We evaluate our video description in the context of action classification with a bag-of-features approach. Experimental results show a significant improvement over the state of the art on four datasets of varying difficulty, i.e. KTH, YouTube, Hollywood2 and UCF sports.

/pdf/action-recognition-by-dense-trajectories-546aqubq3w.pdf

Action recognition by dense trajectories

In this work, we present a novel local descriptor for video sequences. The proposed descriptor is based on histograms of oriented 3D spatio-temporal gradients. Our contribution is four-fold. (i) To compute 3D gradients for arbitrary scales, we develop a memory-efficient algorithm based on integral videos. (ii) We propose a generic 3D orientation quantization which is based on regular polyhedrons. (iii) We perform an in-depth evaluation of all descriptor parameters and optimize them for action recognition. (iv) We apply our descriptor to various action datasets (KTH, Weizmann, Hollywood) and show that we outperform the state-of-the-art.

/pdf/a-spatio-temporal-descriptor-based-on-3d-gradients-2xfqsapd8b.pdf

A Spatio-Temporal Descriptor Based on 3D-Gradients

This paper introduces a video representation based on dense trajectories and motion boundary descriptors. Trajectories capture the local motion information of the video. A dense representation guarantees a good coverage of foreground motion as well as of the surrounding context. A state-of-the-art optical flow algorithm enables a robust and efficient extraction of dense trajectories. As descriptors we extract features aligned with the trajectories to characterize shape (point coordinates), appearance (histograms of oriented gradients) and motion (histograms of optical flow). Additionally, we introduce a descriptor based on motion boundary histograms (MBH) which rely on differential optical flow. The MBH descriptor shows to consistently outperform other state-of-the-art descriptors, in particular on real-world videos that contain a significant amount of camera motion. We evaluate our video representation in the context of action classification on nine datasets, namely KTH, YouTube, Hollywood2, UCF sports, IXMAS, UIUC, Olympic Sports, UCF50 and HMDB51. On all datasets our approach outperforms current state-of-the-art results.

/pdf/dense-trajectories-and-motion-boundary-descriptors-for-48yfw37sv2.pdf

Dense Trajectories and Motion Boundary Descriptors for Action Recognition

Local space-time features have recently become a popular video representation for action recognition. Several methods for feature localization and description have been proposed in the literature and promising recognition results were demonstrated for a number of action classes. The comparison of existing methods, however, is often limited given the different experimental settings used. The purpose of this paper is to evaluate and compare previously proposed space-time features in a common experimental setup. In particular, we consider four different feature detectors and six local feature descriptors and use a standard bag-of-features SVM approach for action recognition. We investigate the performance of these methods on a total of 25 action classes distributed over three datasets with varying difficulty. Among interesting conclusions, we demonstrate that regular sampling of space-time features consistently outperforms all tested space-time interest point detectors for human actions in realistic settings. We also demonstrate a consistent ranking for the majority of methods over different datasets and discuss their advantages and limitations.

/pdf/evaluation-of-local-spatio-temporal-features-for-action-1wwkid1mvv.pdf

Evaluation of local spatio-temporal features for action recognition

We propose a novel human-centric approach to detect and localize human actions in challenging video data, such as Hollywood movies. Our goal is to localize actions in time through the video and spatially in each frame. We achieve this by first obtaining generic spatio-temporal human tracks and then detecting specific actions within these using a sliding window classifier.

We make the following contributions: (i) We show that splitting the action localization task into spatial and temporal search leads to an efficient localization algorithm where generic human tracks can be reused to recognize multiple human actions; (ii) We develop a human detector and tracker which is able to cope with a wide range of postures, articulations, motions and camera viewpoints. The tracker includes detection interpolation and a principled classification stage to suppress false positive tracks; (iii) We propose a track-aligned 3D-HOG action representation, investigate its parameters, and show that action localization benefits from using tracks; and (iv) We introduce a new action localization dataset based on Hollywood movies.

Results are presented on a number of real-world movies with crowded, dynamic environment, partial occlusion and cluttered background. On the Coffee&Cigarettes dataset we significantly improve over the state of the art. Furthermore, we obtain excellent results on the new Hollywood---Localization dataset.

/pdf/human-focused-action-localization-in-video-sasr4mzpst.pdf

Alexander Kläser

Papers

Action recognition by dense trajectories

A Spatio-Temporal Descriptor Based on 3D-Gradients

Dense Trajectories and Motion Boundary Descriptors for Action Recognition

Evaluation of local spatio-temporal features for action recognition

Human focused action localization in video