Re-implementation of the paper Deep Residual Learning for Image Recognition by Kaiming He et Al. using Google's TensorFlow library.

I implemented this paper for my Bachelor's thesis at Technische Universität München, Chair of Scientific Computing.

You can read all about my experiments and results in my thesis (located in the thesis folder). Here's a short summary of the most important findings.

Due to hardware restrictions I was only able to test the models ResNet-20, ResNet-32, ResNet-44 and ResNet-56 on the CIFAR-10 dataset. This is what I got, compared to the results from the original paper:

If you'd like to run your own experiments with my implementation, take a look into the shellscripts folder for examples how to run training and evaluation on different models. All configurable parameters can be found in the file config.py.

On master there also exist stubs for training and evaluating the models on other datasets, namely the ImageNet dataset and the image data of the Yelp Restaurant Photo Classification challenge. For the latter there are working implementations in the branches yelp, yelp-evaluation and yelp-testing. However, I highly discourage using them since the code is quite messy.