AI chief architect 9- huxiaoguang, propeller model library and industry application

0、 Introduce

Catalog

1、 summary

Difficulties in industrial implementation

Propeller model library solution

Panorama of industry and model base

AI Science component

Sample data + Fitting of formula combination

Application cases of fluid mechanics

Performance and accuracy PP Series model

Training reasoning Integration , Transformation of other models

Compression tool

Automatic compression function , Just data and models , The code does not need to be given , Automatic generation

2、PaddleClas

Classification task

Basic tasks

Attribute detection classification

PULC Lightweight solutions

Preset common models

Server side GPU And mobile CPU Network structure ,3X3 Convolution fit GPU

The network model is complete

PP-Shitu Suitable for scenarios where the category is not determined in advance

3、PaddleDetection

Target detection task , The goal is + Location

Facing the challenge

Application scenarios

Basic algorithm , A stage , Two stages , and Anchor Free

Comparison of detection algorithms

Server and mobile models

PP-YOLO

Mobile terminal model

Key point detection

Lightweight detection

application

4、PaddleOCR

Application scenarios

Panorama

Optimization of detection and identification

contrast

5、PaddleNLP

Development Kit

Development experience

General information extraction UIE, Statistical learning , neural network , Preliminary training ,Prompt Time

Model migration capability

Characteristic model

High performance production acceleration library

Dialog generation model

6、 Industry Applications

Intelligent scenes

Enterprise case

scene

background

End side and edge side synchronous inference detection

difficulty

Model and efficiency

tailoring

Distillation

quantitative

Final effect

Actual case

QA Understand the principles behind any contact tool

aistudio In fact, it is through the front end or the client side of the code , Visit the virtual environment of Baidu's server