Copied from the report.
The work is uploaded to Github at esbenkc/tinyimage-classification and the process to run it is simple. Clone the project and run python main.py (optionally run pip install -r ./requirements.txt but this will install a lot of packages). If you wish to see it in Tensorboard, run the command tensorboard --logdir logs/fit which will also show the runs from this report. This will reproduce the best result reported in this report, i.e. the MobileNetV2 transfer learning and fine-tuned model. This will generate the file submission.csv and cnn_mobilenet_model.h5.
The device used is a Windows 11 laptop with 32GB RAM and a CUDA GTX1650Ti with 6GB dedicated memory. To run the models on the GPU, we use CuDNN with Tensorflow GPU.