Subjects / Topics / Fields

▶ Action Recognition

Click Here to Show All

[Definition] Action recognition is to classify short video clips that have been pre-segmented (Actually, videos in real environments are generally not pre-segmented and contain a lot of irrelevant information).
[Supplement] The category of action recognition is usually all kinds of human actions, so it can also be called Human Action Recognition. However, most of the algorithms developed based on this problem are not specific to people, and can also be used for other types of video classification. In addition, the task of action recognition generally does not include the problem of Temporal Action Detection in videos.

Click Here to Show All

[Definition] Affective computing is to establish a harmonious human-computer environment by giving the computer the ability to recognize, understand, express and adapt to human affections, and make the computer have higher and comprehensive intelligence.
[Supplement] In computer vision field, the current affective computing mainly refers to Facial Expression Recognition, which includes three directions: universal expression recognition, Facial Action Coding System (FACS), and continuous expression model Valence-Arousal. In addition, the input of accurate affective computing should be multimodal, including facial expression, voice and intonation, text information, body behavior and even electroencephalogram (EEG) signal. The definition of affection is always ambiguous for its multi-interdisciplinary character. So affective computing is rather difficult and has a long way to go.

Click Here to Show All

[Definition] Here, we will collect some algorithms and materials about people face and head including detection, estimation, reconstruction and recognition in Computer Vision. The input is mostly a 2D image (may also having RGB-D or RGB-times). The outputs are points, bounding-boxes, Euler angles and so on.
[Supplement] Various tasks in deep learning based Computer Vision have strong relation with person's face, sometimes the head. Therefore, we do not discuss them separately. Especially, the face related works or fileds include Face Detection, Face Alignment, Face Reconstruction (3D), Face Recognition, Beautify Face and so on. The head related tasks are mainly about Head Detector and Head/Hand Pose Estimation. We will also introduce some other interesting related topics as long as we have found thems.

Click Here to Show All

[Definition] Model compression is to minimize the consumption of storage space, computing space and time of deep models, and is also committed to accelerating the training and inference of the model.
[Supplement] Deep learning makes the performance of many computer vision tasks reach an unprecedented height. Although the complex model has better performance, the high storage space and computing resource consumption are the important reasons that make it difficult to effectively apply in various hardware platforms. Therefore, model compression is essential. To solve these problems, it is necessary to cut in from many aspects, including machine learning algorithm, optimization theory, computer architecture, data compression, index compilation and hardware design. The methods of model compression can be roughly divided into: low rank approximation, network pruning, network quantification, knowledge distillation and compact network design. This is the last battlefield of large-scale application of computer vision.

Click Here to Show All

[Definition] The task of object detection includes the location and classification of the pre-defined objects in the image.
[Supplement] Object detection is always one of the basic, popular and important tasks in the field of computer vision. Especially after the great breakthrough of deep learning technology in Image Classification task (in 2012), the development of object detection is rather rapid, even now it is still the most active topic. Many technologies, ideas and innovations based on object detection have great significance for reference and promotion in other fields. Object detection is absolutely a bright pearl in the CV crown.

Click Here to Show All

[Definition] Given a 2D RGB image, the traditional task of pose estimation is to predict and output all the keypoints of human bodies contained in it, and connect them into independent skeletons.
[Supplement] On this basis, there are other pose estimation tasks under different types of input. Such as 3D Pose Estimation (RGBD), video based Pose Estimation and Tracking (videos), WiFi Reflection Based Pose Estimation (WiFi info) and Human Body Pose and Shape Estimation (3D motion data). Pose estimation can be used not only for practical applications directly, but also to tackle other computer vision tasks.

Click Here to Show All

[Definition] Visual tracking refers to the detection, extraction, recognition and tracking of moving objects in image sequences.
[Supplement] The purpose of visual tracking is to obtain the motion parameters of the moving object, such as position, velocity, acceleration and trajectory, so as to process and analyze the next step, realize the behavior understanding of the moving object, and complete the higher level visual task. Visual tracking can be divided into two branches, CNN and correlation filtering. At present, CNN has not fully occupied the field of tracking. On the one hand, CNN model has poor real-time processing of images, on the other hand, there is a lot of redundant information in video. Visual tracking is the most significant and challenging task in the CV field.

Name		Name	Last commit message	Last commit date
Latest commit History 2,091 Commits
BasicTech		BasicTech
action_recognition		action_recognition
affective_computing		affective_computing
face_head_related		face_head_related
model_compression		model_compression
object_detection		object_detection
others		others
pose_estimation		pose_estimation
robot_manipulation		robot_manipulation
transfer_learning		transfer_learning
vision_transformer		vision_transformer
visual_tracking		visual_tracking
Field_Elite.md		Field_Elite.md
README.md		README.md
licenses.txt		licenses.txt