Participants:

• Alexis Baladón
• Ignacio Viscardi
• Facundo Pedreira

Libraries

• [scikit-learn] - For dataset splitting, models, hyperparameter tuning and evaluation.
• [tensorflow] - Deep Learning library for model training and utilities.
• [transformers] - Deep Learning library for pretrained models.
• [OpenCV] - For computer vision features.
• [pandas] - Library for structured data manipulation.
• [matplotlib] - Plot functionalities for data analysis
• [seaborn] - Wrapper around matplotlib to make the analysis faster.

Features:

Detection:

Modules

• Exploratory Data Analysis (EDA) for yacine train dataset.
• Implementation of a training pipeline to train a grid of models and parameters.
• Implementation of a predict pipeline to predict and evaluate a dataset given a trained model.

Data & Methods

• Creation of dataset using Zero-Shot clip-vit-base-patch32 model.
• Cache of feature extraction in order to save runtime.
• Data augmentation
• Cross validation which was now migrated to a holdout dataset.

Segmentation:

• Swimming Pool segmentation using K-means on RGB values, erosion + dilation (opening) and HSV information for blue regions.
• Experiments with different methods such as otsu thresholding.

Application:

• Application of both functionalities using satellite images from Uruguay.

Webpage:

The presentation webpage was developed using Angular and is hosted in:

https://alexisbaladon.pages.fing.edu.uy/timag-presentation

Datasets:

The datasets used in the project are:

• Pool-dectection: https://github.com/yacine-benbaccar/Pool-Detection/tree/master
• Algarve's Dataset: https://www.kaggle.com/datasets/cici118/swimming-pool-detection-algarves-landscape
• Unidad de Información Geográfica: https://intgis.montevideo.gub.uy/sit/mapserv/data/fotos_2021/J-29-C-5-N-3.jpg

Program Instructions:

To install dependencies, it is necessary to run the following command:

pip install -r requirements.txt

This software is divided in multiple notebooks and a main program.

This is the description of each directory:

• data: Contains the datasets and cache.
• logs: Contains a register of prorgam executions.
• models: Contains the model hyperparameter grid and the best model binary file.
• notebooks: Contains notebook from each 3 stages of the final solution: Classification, Segmentation and application.
• results: Contains tables with results of classification and segmentation.
• scripts: Contains the script used to change the format of the Algarve's Dataset to the expected project format, and the notebook used to fragment and classify each fragment of the dataset. Aditionally, there are scripts to train the optimal and baseline model.
• src: Contains the main training program which was built using a pipeline architecture (ingestion, transformation, training and evaluation).

Training Script:

To reproduce the training tests implemented in src, the following commands will be of aid:

• Show flags description:

'python main.py --help

• Train the model in the default training dataset (yacine)

'python main.py --train

• Evaluate the trained model using the default evaluation dataset (fragmented Algarve's)

'python main.py --predict

• Use cached features:

py main.py --train --cache_features

• Train with a small and fast grid:

py main.py --train --small_grid