Where to find a reliable tutorial on "Strategies for training a model for image recognition"

General Discussion
,
#1

Dear coder,

I am interested in a tutorial on how to setup layers for image recognition.

Background to this:
I have to pass +63 percent accuracy on recognize if a given image (150px | 150px) shows a cat or a dog. Running thru the provided tutorial-plan gave me information on how to setup the model, train it and how to use predict/evaluate but I have really no idea how to setup the layers to get a good result.

Can someone tell me where I can find tutorials/information about what I have to do (Strategies on building a stack of layer)?

Thank you!

#3

Ahoi Thorsten,

sounds like you’re searching for a CNN (Convolutional Neural Network) Model & Architecture.
https://developers.google.com/machine-learning/glossary/#convolutional_layer
https://www.tensorflow.org/tutorials/images/cnn

For more insights and lectures, feel free to have a deep dive here:
https://cs231n.github.io/convolutional-networks/
https://www.youtube.com/watch?v=iaSUYvmCekI

Interesting Codelab Tutorials:
Horse or Human (Python):
https://developers.google.com/codelabs/tensorflow-5-compleximages

Optimising CNN’s (Python):
https://developers.google.com/codelabs/tensorflow-4-cnns

MNIST with TF.js (JavaScript)
https://codelabs.developers.google.com/codelabs/tfjs-training-classfication/index.html#4

Some, already setup, browser based playground(s), e.g for visualisation purpose:
https://cs.stanford.edu/people/karpathy/convnetjs/demo/cifar10.html
https://cs.stanford.edu/people/karpathy/convnetjs/demo/mnist.html
(hmm, somehow i couldn’t find the tf.js playgrounds adhoc anymore …)

Maxpooled,
Dennis

2 Likes