fchollet/deep-learning-models

★ 7,350PythonAudience · researcherComplexity · 2/5LicenseSetup · easy

Mindmap

mindmap
  root((deep-learning-models))
    Status
      Deprecated
      Use keras.applications
    Models included
      VGG16 and VGG19
      ResNet50
      Inception v3
      Music tagging RNN
    What they do
      Classify images
      Extract visual features
      Identify objects in photos
    Tech
      Python
      Keras
      TensorFlow or Theano
    Audience
      ML researchers
      Data scientists

mindmap root((deep-learning-models)) Status Deprecated Use keras.applications Models included VGG16 and VGG19 ResNet50 Inception v3 Music tagging RNN What they do Classify images Extract visual features Identify objects in photos Tech Python Keras TensorFlow or Theano Audience ML researchers Data scientists

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Pass a photo to a pre-trained model and get back a plain-English label describing what is in the image (e.g. 'African elephant').

USE CASE 2

Extract numerical feature representations from an image using an intermediate network layer to feed into your own custom classifier.

USE CASE 3

Use a pre-trained model as a starting point for fine-tuning on your own image dataset without training from scratch.

Tech stack

PythonKerasTensorFlowTheano

Getting it running

Difficulty · easy Time to first run · 30min

This repo is deprecated, use keras.applications instead, pre-trained weights were downloaded automatically from the internet on first load.

Code is MIT licensed, use freely for any purpose including commercial, as long as you keep the copyright notice. VGG weights are Creative Commons Attribution, ResNet and Inception weights are MIT.

In plain English

This repository is deprecated. The README states explicitly that it should no longer be used and directs anyone looking for this functionality to the keras.applications module instead. Pull requests are not reviewed and issues are not monitored. While it was active, this repository provided Python code and pre-trained weights for several well-known image classification models built on Keras. The models included VGG16, VGG19, ResNet50, Inception v3, and a recurrent neural network for music tagging. These model names refer to neural network architectures that were trained on large image datasets and can recognize objects in photos. The models were compatible with both TensorFlow and Theano, two earlier backends for Keras. When you loaded a model, it would automatically download the pre-trained weights from the internet and store them in a local cache folder. From there you could pass a photo to the model and receive a prediction of what the photo contains, or use the model to extract numerical representations of images (called features) for use in your own project. The examples in the README show three uses: loading a photo and getting a label back (such as "African elephant"), extracting features from the top layers of the network, and extracting features from a specific intermediate layer inside the network. Each use case requires only a few lines of Python. The code is under the MIT license. The VGG model weights came from Oxford's VGG research group under a Creative Commons Attribution license, and the ResNet weights came from a researcher at Microsoft Research under MIT. The Inception v3 weights were trained independently and released under the MIT license as well.

Copy-paste prompts

Prompt 1

I want to classify images using a pre-trained ResNet50 model in Keras. Show me the modern keras.applications equivalent of what fchollet/deep-learning-models used to do, with code to load a photo and get the top-5 predicted labels.

Prompt 2

Using the Keras keras.applications module (the replacement for fchollet/deep-learning-models), show me how to extract feature vectors from the second-to-last layer of VGG16 for a folder of images so I can use them for similarity search.

Prompt 3

Help me fine-tune an Inception v3 model from keras.applications on my own dataset of 500 photos with two classes. Show the full training loop with data augmentation.

Prompt 4

Explain the difference between VGG16, ResNet50, and Inception v3 in plain English, and tell me which one to use for a project where I need fast inference on a laptop CPU.

Open on GitHub → Explain another repo

← fchollet on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.