Machine Learning ย - part 1- Research ย Gathering info
In this post I am going to summarize and bring sources to start understanding practical approach to Machine Learning.ย
You can Follow this series of posts or follow me on Facebook tech page.
The ground truth is what you measured for your target variable for the training and testing examples.ย
Nearly all the time you can safely treat this the same as the label. ย
* Building a model - meaning we need to define the goal of the coding.
For instance recognize where this shirt is from is different from recognize that the man is wearing a shirt.
The geographic location matters - for instance the Europe and USA clothing items and trends will be different,
We need to train a network with new data, for instance images - itโs called - ย Convolutional Neural Network , or shortly CNN.
How Convolutional Neural Networks work
CNN trains with a lot of data .
ืืืืื ืขืืืงื (Deep Learning) ืืื ืขื ืฃ ืฉื ืืืืืช ืืืื ื (Machine Learning), ืืืื ื ืืฉื ืืืื ืื ืืืืื ื.
ืืืื ืืงื ืืื, ืืืืื ืขืืืงื, ืืืืื ืืืืื ืืืฆืืืื ืฉืืืืฉืืื ืฉื ืคืืฆ'ืจืื ืืฉืืจืืช ืืชืื ืชืืื ืืช, ืืงืกืืื ืืงืื โ ืืืจ ืืืงื ื ืื ืืชืจืื ืืช ืจืืื.
ืืืืืื ืื "ื ืืชืืฆืขืช ืืืืฆืขืืช ืฉืืืืฉ ื-Convolutional Neural Network โ ืื ืืงืืฆืืจ โ CNN โ ืฉืื ืกืื ืฉื ืจืฉืช ืขืฆืืืช.
ื-CNN ืืืืื ืช ืขื ืืืกืฃ ืืืื ืฉื ืืืืข, ืืฉืจ ืืขืืจืชื ืืจืฉืช ืืืืืช ืืืฆืืืื ืขืฉืืจืื ืฉื ืคืืฆ'ืจืื, ืืฉืจ ืื ืืืื ืืจืื ืืืงืจืื ืชืืฆืืืช ืืืืืช ืืืชืจ ืืืื ืฉืื ืืืื ืืคืืฆ'ืจืื ืืืกืืจืชืืื.
ื-CNN ืืืืื ืช ืขื ืืืกืฃ ืืืื ืฉื ืืืืข, ืืฉืจ ืืขืืจืชื ืืจืฉืช ืืืืืช ืืืฆืืืื ืขืฉืืจืื ืฉื ืคืืฆ'ืจืื, ืืฉืจ ืื ืืืื ืืจืื ืืืงืจืื ืชืืฆืืืช ืืืืืช ืืืชืจ ืืืื ืฉืื ืืืื ืืคืืฆ'ืจืื ืืืกืืจืชืืื. ย
ืฉืืืืฉ ื-CNN ืืชืืจ ืืืืฅ ืคืืฆโืจืื
ื ื ืื ืฉืื ืื ื ืจืืฆืื ืืกืืื ืชืืื ื ืืืืช ืืืื ืืืคืฆืืืช ืืื ืืื, ืืื, ืืืคื ืืื ืืื'. ืืืืฉื ืืกืื ืืจืืืช ืฉื ืืืืืช ืืืื ื ืืื ืงืืื ืื ืืืืฅ ืคืืฆ'ืจืื ืืขื ืืื ืื ืืชืื ืืชืืื ื โ ืืื edge-ืื, ืคืืืื ืฆืืขืื ืืืืืื โ ืืื ืฉืื ืืืจืื ืืืืื ืฉืืืืืจืืชืื ืืืืื ืืืื ื ืกืื ืืจืืื ืื ืืืืขืื ืืคืขืื ืืฉืืจืืช ืขื ืชืืื ื, ืื ืืชืขืืืื ืืืืืืื ืืืืื ื ืฉื ืชืืื ื. ืืฉืื ืืื โ ืืชืืฆืข ืืกืืืื ืฉื ืืชืืื ื, ืืขืืจืช ืืกืืื ืืฉืจ ื ืื ื ืงืืื ืืื ืขื ืกืื ืชืืื ืืช ืืืืืื.
ืืืืฉื ืฉื ืืืืื ืขืืืงื, ืืขืืืช ืืืช, ื ืืชื ืื ืืืืืืจืืชื ืืืืื ืืช ืืคืืฆ'ืจืื ืืืืืืืืช ืืชืื ืืชืืื ืืช, ืืคืืฆ'ืจืื Low Level ืื ืจืืื ืืื edge-ืื ืืคืื ืืช, ืืขื ืืคืืฆ'ืจืื ืกืคืฆืืคืืื ืืืขืื. ืืืืืจ, ืืืืืจืืชืื ืืืืืื ืืขืืืงื ืื ืืืืขืื ืืืฆืข ืจืง ืืช ืืกืืืื, ืืื ืื ืื ืืืืขืื ืืืืื ืืืฆื ืืืืฅ ืคืืฆ'ืจืื ืืฉืืจืืช ืืชืื ืืชืืื ืืช, ืืื ืื ืืืกืืื ืืช ืืฆืืจื ืืืืืืฅ ืืื ื ืฉื ืืคืืฆ'ืจืื, ืืืขืฆื ืืืืฉืื ืืืืื ืืงืฆื ืืงืฆื (End to End).
* We need to have a scope of images with and the algorithm will learn from it to recognize the items. For instance - http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html ย
Category and Attribute Prediction Benchmark evaluates the performance of clothing category and attribute prediction
In-shop Clothes Retrieval Benchmark evaluates the performance of in-shop Clothes Retrievel. This is a large subset of DeepFashion, containing large pose and scale variations. It also has large diversities, large quantities, and rich annotations,
Consumer-to-shop Clothes Retrieval Benchmark evaluates the performance of consumer-to-shop Clothes Retrievel. This is a large subset of DeepFashion, containing cross-domain correspondences and variations in the wild.
Fashion Landmark Detection Benchmark evaluates the performance of fashion landmark detection. This is a large subset of DeepFashion, with diverse and large pose/zoom-in variations.
We can donwload and use those images to train our algrithm in many capacities..
Trainig data is pretty well here.
For more details of the benchmarks, please refer to the paper,
DeepFashion: Powering Robust Clothes Recognition and Retrieval with Rich Annotations, CVPR 2016.
1. Category and Attribute Prediction Benchmark:
Category and Attribute Prediction Benchmark
2. In-shop Clothes Retrieval Benchmark:
In-shop Clothes Retrieval Benchmark
3. Consumer-to-shop Clothes Retrieval Benchmark:
Consumer-to-shop Clothes Retrieval Benchmark
4. Fashion Landmark Detection Benchmark:
Fashion Landmark Detection Benchmark
ืืืื PYTHON, KERAS" TENSOR
Suggested way to go - install Anacodna which will includes the relevant software.
https://docs.continuum.io/anaconda/install/windows
https://medium.com/learning-machine-learning/getting-tensorflow-theano-and-keras-on-windows-70c18f2c533b
So if we want to match styles and products together , Indico company made a limited test to make a suggestion and matching between items - like top and pants. Although they talk about the limitness of their model , if we would take an individual this model could work for day in day out adviser of what actually to put on before we leave for work/event etc..
https://www.youtube.com/watch?v=Ikg1xNdabFo&t=180s
Tensor Flow - ย Build a TensorFlow Image Classifier in 5 Min
https://www.youtube.com/watch?v=QfNvhPx5Px8
How to Retrain Inception's Final Layer for New Categoriesย