85 datasets found

The SHG-VQA model predicts a situation hyper-graph structure composed of existing actions and relations in the input video.

This paper proposes a method for action search in videos, which is used for spotting actions in videos.

This paper proposes a method for temporal action localization using single-frame supervision. The authors use a unified system called SF-Net to predict actionness scores and...

The Breakfast dataset is another dataset used in the paper, which contains 712 videos of people performing various activities, such as making coffee or scrambling eggs. The...

The dataset used for unsupervised image-to-image translation, consisting of a set of 20 possible actions performed by 10 different human renders.

A large-scale labeled RGB-Depth-Optical Flow dataset for action recognition, containing 1,074 long untrimmed videos with paired RGB, depth, and optical flow modalities for 51...

AViD is a new public video dataset for action recognition, containing action videos from diverse countries.

Skeleton-based action recognition has recently received considerable attention. Current approaches to skeleton-based action recognition are typically formulated as one-hot...

CATER (a dataset for Compositional Actions and TEmporal Reasoning) was released by [45] under the Apache 2.0 license.

The FaMoS dataset records 95 subjects, each performing 28 predefined actions (e.g. anger, disgust, fear, surprise).

VOCA-2012 is an action recognition dataset with task-dependent eye-fixation data.

Recognizing and comprehending human actions and gestures is a crucial perception requirement for robots to interact with humans and carry out tasks in diverse domains, including...

The THUMOS'13 dataset contains 24 classes and 3,207 videos.

The JHMDB dataset contains a varying number of actions (21 in JHMDB) across different domains (sports and daily activities).

The UCF Sports dataset contains a varying number of actions (10 in UCF Sports, 21 in JHMDB, and 24 in THUMOS’13) across different domains (sports and daily activities).

Kinetics Skeleton 400 is a dataset adapted from the Kinetics 400 video dataset using the OpenPose toolbox in 2D keypoint modality.

Temporal action detection in untrimmed videos via multi-stage cnns, Cdc: convolutional-de-convolutional networks for precise temporal action localization, Temporal action...

N-UCLA dataset is a widely used skeleton-based action recognition dataset, containing 1494 video clips featuring 10 volunteers.

The dataset used in the paper is a collection of videos and corresponding referring expressions.

Human pose estimation refers to the task of recognizing postures by localizing body keypoints (head, shoulders, elbows, wrists, knees, ankles, etc.) from images.