ML4CV---Assignment

The notebook contains the assignment for the course of Machine Learning for Computer Vision at the University of Bologna. The notebook has been created using Kaggle. All the model training and evaluation have been done with the Kaggle GPU P100. To ensure reproducibility please upload the notebook on Kaggle, connect to GPU P100 and run the notebook. Please, ensure that on your Kaggle notebook internet connection is enabled. It should automatically pin the environment to the original ones, do not change this otherwise some packages may give some errors (Kaggle environment changes often).

Dataset and Data

Kaggle is fully cloud-based. Your local machine and the Kaggle server don’t share a file system, this implies that all the files need to be dinamically downloaded during the notebook execution, or manually added. To avoid this, i proceded as follows:

• The StreetHazards dataset has been directly downloaded from the Berkley university link.

Structure

The notebook is divided in two main parts:

• the first part contains all the functions and classes used during the task documented;
• the second part contains the task, where the results and ablation have been showed and commented.

Configuration

The notebook take the results of the evaluation, the weights and the logs directly from the results directory provided in the Google Drive. It comes in a configuration where only the Qualitative results are runned in real time, all the other results are stored in the results directory. You could alternatively decide to:

• Run the Evaluation set the variable EVALUATION to True in the section global variables. If you decide to run the EVALUATION only for some configuration in the ablation you have to run, at least, the cell of the best segmenter and the best rpl, otherwise running any other cell of the ablation will arise a key error. This happens because in the ablation comparisons with the best model is showed and the structure inside the results directory and the one computed at run time does not match. This problems arises from Kaggle notebook whihc is thinked to be runned in sequence. Of course, if you run all the cells everything is fine;
• Train the best segmenter model (may require a bit) set the variable TRAIN_MODEL to True in the section global variables;
• Compute the mean and standard deviation set the variable COMPUTE_MEAN_STD to True in the section gloabal variables.