Utility functions for the practical assignments of the courses:
-
Deep Learning for Image and Video Processing: Learning Strategies and Applications
(MUDLAVai, Universidad Autónoma de Madrid)
https://www.uam.es/uam/master-universitario-aprendizaje-profundo-tratamiento-senales-audio-video -
Deep Learning for Visual Signal Processing
(IPCVAI, Universidad Autónoma de Madrid)
https://ipcv.eu
Author: Juan Carlos San Miguel (📧 juancarlos [dot] sanmiguel [at] uam [dot] es)
🌐 http://www-vpu.eps.uam.es/jcsanmiguel
dlvsp-utils is a lightweight utility package used across the course notebooks.
It provides reusable helpers for:
- dataset manipulation and inspection
- accuracy computation and per-class reporting
- simple visualization utilities for model analysis
The goal is to reduce boilerplate code and keep the focus on learning strategies and experimental analysis.
Install directly from GitHub:
pip install git+https://github.com/jcsma/dlvsp-utils.gitfrom torchvision import datasets, transforms
from dlvsp_utils.data import select_classes_dataset, inspect_dataset_classes
from dlvsp_utils.metrics import calculate_accuracy, print_accuracy_report
train_full = datasets.CIFAR10(root="./data", train=True, download=True, transform=none)
train_ds, class_names = select_classes_dataset(train_full, ['cat','dog'])
inspect_dataset_classes(train_ds, class_names=class_names, header="\nTRAIN:")The repository contains the following modules:
src/dlvsp_utils/data.pyDataset utilities (class selection, inspection, sampling helpers)src/dlvsp_utils/metrics.pyAccuracy computation and per-class performance reportingsrc/dlvsp_utils/viz.pyVisualization helpers for analysis and debuggingpyproject.tomlPackage configuration and dependencies