Skip to content

Supporting scripts to evaluate different TensorFlow/Keras models on ImageNet (ILSCRV2012) classification dataset.

License

Notifications You must be signed in to change notification settings

nrasadi/imagenet_validation

 
 

Repository files navigation

How to reproduce ImageNet validation results

This code in this repository can be used to reproduce the ImageNet validation results for Keras pretrained models. A blog post describing this process in more detail is here.

Setup instructions

The ILSVRC2012 validation images should be put in data/val/. The remaining files mentioned in the blog post, and enumerated below, should be put in data/:

  • ILSVRC2012_validation_ground_truth.txt
  • meta.mat
  • synset_words.txt

After the data is organized the notebooks can be run in order to reproduce model validation results.

=======

Update 2/4/2021

  • I was unable to find the ILSVRC2012 devkit online, so have included a copy of the 3 files (mentioned above) needed to run this in data/ as well as the COPYING notice from the devkit here.
  • I was able to reproduce these results with tensorflow=2.3.1 and keras=2.4.3 (see notebook 2)

About

Supporting scripts to evaluate different TensorFlow/Keras models on ImageNet (ILSCRV2012) classification dataset.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 100.0%