TRS IS-CV·Computer Vision Software

TRS IS-CV is a software product for searching for similar content fragments between images and videos. It integrates deep learning frameworks, including Tensorflow and PyTorch. It relies on key technologies such as image/video retrieval and classification, duplicate image/video detection, image/video object detection and recognition, image/video OCR text extraction and scene classification, the deep learning model is used to understand images/videos, extract image/video content features, and establish an image/video retrieval system. By inputting images/videos, users can quickly retrieve images/video clips similarity with input video in custom image/video retrieval database, and effectively avoid the impact of image/video format conversion, editing, clip splicing, compression, rotation and other transformations on search results.

TRS IS-CV can identify and screen out illegal data such as pornography, violence, and politically sensitive data from massive image/clip video data in real time, as well as illegal data such as advertising push, bad taste, and vulgarity, to ensure the security and quality of video content on the Internet. It can be widely used in scenarios such as copyright identification, advertising tracking, content review and de-duplication.

Product Advantages

Complete functions, highly integrated system

Focusing on the three core technologies of computer vision,intelligent speech processing and multimodal retrieval.It integrates a variety of algorithm models that have been trained and tuned to provide users with high-quality AI service capabilities, and provides more than 100 large + small models of machine learning and deep learning algorithms such as text, images, and audio and video. It provides full-stack AI technology capabilities, dozens of APIs, and SDK development mode.

Landing Scene

Combined with TRS company's AI+ strategy, it fully unleashes the power of big data by combining it with industry data such as media, finance, public opinion and security, and realizes the application model of "AI+ industry scenarios" in various industries.

Privatized & Customized

The software supports privatized deployment and implementation to ensure the data security of industry users, and provides training interfaces and training tools, which can customize labeled datasets and develop industry models for the domain.

Support of ITAI (Information technology application innovation)

We support ITAI software and hardware products, including operating systems such as Kirin, UnionTech, and Deepin, CPUs such as Feiteng, Kunpeng, Loongson, and GPU products such as Huawei Ascend and Haiguang DCU. We support the integration of hardware of different architectures such as Intel, Nvidia, Feiteng, and Loongson into a unified computing service framework. We encourage users using GPUs and NPUs to serve the computing-intensive applications of specific AI models, using CPUs to serve general applications according to the computing characteristics of different applications and models.

Product Functions

OCR

Based on a variety of algorithms, it detects and recognizes text in different scenes, suitable for multiple scenarios like taking pictures, scanning documents, handwriting, and natural scenes, as well as providing text recognition for different lengths, fonts, and languages. Support table structure recognition, seal recognition, etc.

Face Recognition

This function can help users identify important and sensitive people in images or videos, such as major national leaders, well-known artists, and bad actors, and then use the obtained face information to mark the material and generate an overall description. Self-built face database is supported. It can be widely used in business scenarios such as image and short video content review.

Image Recognition and Scene Understanding

Based on deep learning algorithms, it can accurately identify the important information contained in images, which can be applied to image classification, object detection, scene recognition and other scenarios. It provides image recognition in professional fields based on industry knowledge.

Audio Processing

This feature makes it easier for users to mark audio data and convert audio to text, efficiently and accurately extract behaviors, tags, events and other information from audio. It includes speech recognition, speech generation, and speech classification functions. It can be widely used in business scenarios such as intelligent video recognition.

Multi-Modal Retrieval

We achieve large-scale cross-modal and multi-modal retrieval by using multi-dimensional and large-scale annotation of pictures and videos, we extract text and visual information from Embedding multi-dimensional space, helping users efficiently and accurately obtain the theme, layout, entity, event and other information related to the retrieved materials. It can be widely used in business scenarios such as image and video comparison and duplicate checking。

Content identification capabilities

We using multi-modal analysis technologies on text, images, audios and videos, it can automatically detect content related to pornography, advertisements, terrorism and violence, sensitive people, etc., to help customers reduce the risk of business violations.

Image-video Retrieval

We provide multi-modal large model image semantics, image feature extraction/analysis and other technologies, to carry out the deep genetic coding calculation of images, search technologies such as image-text mixed indexing are used to help customers find images and videos from the database by combining different application services and industry scenarios.

Application Scenario

Project of Bank Consumer Protection

Postal Savings Bank Annual Report Analysis Project

Customs Information Center Project

Multi-Modal Retrieval Project

TRS IS-CV provides users OCR capability and content analysis functions based on deep learning models, realizes compliance review of bank slogans, financial terms, financial indicators, etc., we provides model training and evaluation functions.

TRS IS-CV provides users large-scale PDF analysis, table structure analysis and OCR recognition functions based on deep learning models, achieve text extraction and layout restoration of PDFs with different layout styles of various banks, helps users quickly extract key information of annual reports.

TRS IS-CV provides users with image retrieval and comparison functions based on deep learning framework, building the functions of frozen product label retrieval and comparison, harmful insect detection, wood detection, and provides model training and evaluation functions.

TRS IS-CV provides users with text/image/video retrieval and comparison functions based on cross-modal deep learning frameworks, achieve multi-modal retrieval functions based on people, scenes, and long texts.