Machine Learning hackathon held at LTU. This was initialized for the group that trains networks on rectangular images.
Use resnet/vgg/alexnet that have been trained on IMAGENET using rectangular images. These pre-trained models will then be used to classify images of documents from the Tobacco3482 dataset. The idea is to observe the original aspect ration as much as possible for the documents (rectangular) to loose as little information as possible.
Run with commant:
python run-benchmark.py 2>&1 | tee -a outputs/$(date +%Y%m%d_%H%M%S).log
This will save the output in a file formatted as YYYYMMDD_hhmmss.log in the outputs folder.