3. Getting Started with Pre-trained Models on ImageNet

ImageNet is a large labeled dataset of real-world images. It is one of the most widely used dataset in latest computer vision research.

imagenet

In this tutorial, we will show how a pre-trained neural network classifies real world images.

For your convenience, we provide a script that loads a pre-trained ResNet50_v2 model, and classifies an input image. For a list of all models we have, please visit Gluon Model Zoo.

Demo

A model trained on ImageNet can classify images into 1000 classes, this makes it much more powerful than the one we showed in the CIFAR10 demo.

Download demo_imagenet.py

With this script, you can load a pre-trained model and classify any image you have.

Let’s test with the photo of Mt. Baker again.

image0

python demo_imagenet.py --model ResNet50_v2 --input-pic mt_baker.jpg

And the model predicts that

The input picture is classified to be
    [volcano], with probability 0.558.
    [alp], with probability 0.398.
    [valley], with probability 0.018.
    [lakeside], with probability 0.006.
    [mountain_tent], with probability 0.006.

This time it does a good job. Note that we have listed the top five most probable classes, because with 1000 classes the model may not always rank the correct answer highest. Besides top-1 accuracy, we often also consider top-5 accuracy as a measurement of how well a model can predict.

Next Step

If you would like to dive deeper into ImageNet training, feel free to read the next tutorial on ImageNet Training.

Or, if you would like to know how to train a powerful model tailored to your own data, please go ahead and read the tutorial on Transfer learning.

Total running time of the script: ( 0 minutes 0.000 seconds)

Gallery generated by Sphinx-Gallery