Compare ImageNet CNN architectures

Compare ImageNet CNN architectures
09. Aug 2020
1 minute read

Last week, I released a beginner’s utility to compare classification results from SotA Convolutional Neural Network architectures.

Motivation for the tool was to enable beginners and non-programmers to “see” how off-the-shelf pre-trained deep learning models classify their own real-world images. These models wouldn’t necessarily work well across all your images and meant to be more of an educational tool.

Under the hood, it uses pre-trained MobileNetV2, ResNet50, VGG19, InceptionV3, Xception weights from TensorFlow Keras

You can header over here and upload an image or paste the URL to compare classification results.

Here’s are some interesting examples:

  • Calvin and Hobbes image was detected accurately as comic book 💬 by most models 😀

    Calvin and Hobbes

    Calvin and Hobbes

  • Still image of Simba and Zazu from The Lion King movie. InceptionV3 and Xception recognized both 🦁 Lion and 🐦 Hornbill in the top 5 results.

    Simba and Zazu from The Lion King

    Simba and Zazu from The Lion King

Try yourself

Tweet