createamind / cpc-pred-gan Goto Github PK
View Code? Open in Web Editor NEWbase on Representation Learning with Contrastive Predictive Coding and https://github.com/davidtellez/contrastive-predictive-coding
base on Representation Learning with Contrastive Predictive Coding and https://github.com/davidtellez/contrastive-predictive-coding
paper: better prediction imagination condition gan : pattern gan;
学习规律的泛化。基于规律认识的视频预测。学规则,学规律? 用错了数据集在视频预测中?
tech:
--1 more pred item; --1.1 cpc train item pred4 only gen1 1.1.1 cpc train 4-4 val 4-4 and !4-1 !!
--1.1.1 train cpc 4-4 gen 4-1 这个现在就可以!--1.2 gen no noise is better??
--2 gen loss: only cpc loss or what weight?? or 看结果,loss比重一样也行,cpc收敛很快;还是 cpc 100吧
--3 trained weight load; keras tf ok;--3.2 print x item; 可以显示生成数字的cpc判别结果,如果不对再生成!
4 现在训练慢是训练数据是实时生成的,不是预生成的,应该生成好,直接训练
--5 cpc和gan 循环训练,cpc的强化也继续,但是不是用gan的数据训练cpc,是原始cpc的数据训练cpc;gan的训练,Dcpc和Dcirit权重也需要动态变化,先生成好图片,在cpcloss生成准确数字的图片
生成数字的预训练,降低训练生成数字对数字序列的影响。
5 64X64 大数字。
app:
0 not mnist ; 标准数字的数据集;学规律,不是学数字识别
1 more practice; 1.1 123 124; just train ok; 1-500 600-700valid not work;
1.2 1+3=4; 9内加法99内加法
数据是否用标准数字图像,而不是mnist数据集?
1.1.1 数据集生成;
基于之前1234最基本的训练基础上再训练,不是直接记忆序列,从规律开始学习。
个位数训练的顺序情况应用到加1的加法上去;同时训练加法和顺利排列??
或者两种训练同时学习,rnn如何兼顾两种格式?? 即学习个位数的循环排列顺序,也学习数字增长的规律,
或 oneshot;每种数据给一点提示。,数据分段生成。
condition gan ref cgan
1.2.1 数据集生成;
?1 进位规律的学习。
2 picture and video and carla
U-net pix2pix
c3d cdu-net; vid2vid
学规则,学规律? 用错了数据集?
3 video+action ;红绿灯问题
cpchrl
cpc(video+action::video+action paire)
4 rl;强化学习只需要cpc的预测隐变量即可,隐变量生成再可视化跟人交互,
4.2: rl涉及动作。 根据规律,且加入动作进行预测。 规律+动作conditon规律的组合生成隐变量
rnn的数字生成器,生成多个的序列,根据序列个数生成可变序列, rnn的condtion gan;
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.