Download and prepare data for MIT Indoor Classification. The final form of your submission will be a Jupyter Notebook. We suggest that you use a Google Colab. Also, we suggest that you use PyTorch to prepare your ML pipeline. When you are done with it, download your notebook and submit it to us at people@310.ai
Create a minimal baseline for the classification task using 20% of data as validation.Use balanced-accuracy as your metric. Name this notebook “Baseline”. In this part, we focus on your code being clean, simple and understandable. We expect a clear pipeline, going over data loading, model creation, training and evaluation.
Use any tricks you like (ex. model architecture, augmentation, etc.) to improve the metric. The only method you are not allowed to use is pretraining using other datasets (ex. ImageNet pretraining). A minimum balanced-accuracy of 55% is required. Your submission will be examined based on its performance on a random validation (not your own split). The focus on this part is to measure your ability to do critical thinking and come up with good ideas to attack a hard problem. Name this notebook “Challenge” in your submission.