Awesome

Training VGG-16 on ImageNet with TensorFlow 2 and Keras

Description

This is an implementation of the VGG-16 image classification model using TensorFlow 2 and Keras written in Python. The ImageNet dataset is required for training and evaluation. Inference can be performed on any image file.

The code is capable of replicating the results of the original paper by Simonyan and Zisserman. Sample training results can be found in the spreadsheet included in the repository.

References

VGG-16 was initially presented as configuration D in this paper:

Very Deep Convolutional Networks for Large-Scale Image Recognition Karen Simonyan, Andrew Zisserman ICLR 2015

It is available here: https://arxiv.org/abs/1409.1556

The ImageNet dataset must be procured with a valid academic email address from the Stanford Vision Lab: http://www.image-net.org/

Rumor has it that Torrents are available but you didn't hear that from me.

Usage

Make sure you are running at least Python 3.8 and install the required PIP packages listed in requirements.txt. I highly recommend using Anaconda on Windows or a TensorFlow Docker container on Linux.

Once in your Python environment of choice:

pip install -r requirements.txt

At the time of this writing (February 2021), if you are using an Nvidia RTX 30-series GPU, you may need to edit that file and use tf-nightly-gpu for TensorFlow. Otherwise, you might see obvious numerical issues during training.

Run the script from the repository directory:

python -m VGG16

For usage examples, read the comments at the top of __main__.py.

Good luck!