What can target tracking be applied to

Deep learning actually refers to a collection of algorithms that apply a variety of machine learning algorithms on multilayer neural networks to solve problems such as image, text, speech, etc., aiming at extracting more abstract high-level features. In terms of its own structural research, the main ones are:

Convolutional neural networks, recurrent neural networks, and ordinary deep neural networks based on supervised learning; deep generative models based on unsupervised learning (Boltzmann machines, DBN, generative adversarial networks), self-encoders; optimization training, small sample training; transfer learning combined with neural networks; reinforcement learning; graph models, and so on.

From the application field, the main research directions are:

1. Computer vision: image classification, target detection, target tracking, semantic segmentation, face recognition and analysis, image/video search, text recognition, and other directions;

2. Speech recognition, speech enhancement, speech synthesis, etc.

3. Natural language processing, mainly machine translation , text mining, semantic analysis, information retrieval extraction, question and answer system, etc..

There are also deep learning applications in video analytics, big data, robotics direction. At present, deep learning is widely used, high performance, but there are also these several problems, such as: sometimes it is difficult to obtain a large amount of supervisory data or labeling cost is high, how to adapt to the small sample training; how to effectively carry out large-scale parallel computation and training; general specific problems and data, the need to customize the structure of the network, how to study the general deep learning model and so on, these are all the future direction of research.