A Photograph showing a dog.

I'm Lucas, a PhD student in Computer Vision/Machine Learning at University of Campinas. My research interests are Explainability and Weakly Supervised Semantic Segmentation problems.

Blog Posts

Activation, Cross-Entropy and Logits

Discussion around the activation loss functions commonly used in Machine Learning problems.

Santos Dumont Super Computer

Accessing and using the SDumont infrastructure for Deep Learning research.

Segmentation and Detection

A comparison between edge-based segmentation, Felzenszwalb's method and morphological segmentation.

K-Means and Hierarchical Clustering

Efficient clustering algorithms implementations in TensorFlow and NumPy.

The 2D Fourier Transform

Applying the Fourier Transform over images using TensorFlow.

A Vectorized Implementation of The Convolution Operation

Implementing a few numeric operations over images using Python programming language.

Class Activation Mapping

Explaining AI with Grad-CAM.

Explaining Machine Learning Models

Explainability using tree decision visualization, weight composition, and gradient-based saliency maps.

Check all posts

Videos & Presentations

Exploring XAI Methods in ML Problems and Complementary Reg. Strategies in WSSS

My qualifying exam presentation at IC - Unicamp, presented in the year of 2022.

Connoisseur: Provenance Analysis in Paintings (SSCI 2021)

Video presentation for the paper, presented at IEEE SSCI 2021.

MinMax-CAM: Improving Focus of CAM-Based VisualizationTechniques in Multi-Label Problems

Video presentation for the article, presented at VISAPP 2022.

NAO navigation in V-REP environment with QLearning (YouTube)

Q-Learning applied to the navigation problem, using NAO robot and V-REP simulation environment.

Aquarium (YouTube)

Aquarium project, written in Java.

Adversarial Feature Hallucination Networks for FSL of Provenance in Paintings - IA-386

Apresentação do projeto final da disciplina IA368CC da Faculdade de Engenharia Elétrica e de Computação da Universidade Estadual de Campinas, ofertada no segundo semestre de 2021.

Classificando impacto humano sobre a floresta amazônica a partir de imagens de satélite (YouTube)

Apresentação do projeto final da disciplina MC886/MO444 do Instituto de Computação (Unicamp), ofertada no primeiro semestre de 2021.

See all videos on YouTube

Scientific Contributions

Bellow is a short (and unorganized) list of my scientific and technical contributions.

Publications

L. David and H. Pedrini, and Z. Dias. "P-NOC: Adversarial training of CAM generating networks for robust weakly supervised semantic segmentation priors," in Journal of Visual Communication and Image Representation (JVCI), v.102, 2024, pp. 104187

paper arxiv code doi

L. David, L. Liborio, G. Paiva, M. Severo, E. Valle, H. Pedrini, and Z. Dias. "Adversarial Feature Hallucination in a Supervised Contrastive Space for Few-Shot Learning of Provenance in Paintings," in 2023 IEEE Latin American Conference on Computational Intelligence (LA-CCI), 2023, pp. 1-6

paper code doi

L. David, H. Pedrini, and Z. Dias. "MinMax-CAM: Increasing Precision of Explaining Maps by Contrasting Gradient Signals and Regularizing Kernel Usage," in International Joint Conference on Computer Vision, Imaging and Computer Graphics, 2022, pp. 222--247

paper code doi

G. Bianchin, L. David, R. Padilha, A. P. da Silva, F. de Paula, L. Infante, L. Jorge, P. Xavier, and Z. Dias'. "Bias Assessment in Medical Imaging Analysis: A Case Study on Retinal OCT Image Classification.," in ICAART (3), 2022, pp. 574--580

paper

L. David, H. Pedrini, and Z. Dias. "MinMax-CAM: Improving Focus of CAM-based Visualization Techniques in Multi-label Problems," in 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, v.4, 2022, pp. 106--117

paper doi

L. David, H. Pedrini, Z. Dias, and A. Rocha'. "Connoisseur: provenance analysis in paintings," in 2021 IEEE Symposium Series on Computational Intelligence (SSCI), 2021, pp. 1--8

paper code doi

L. David, H. Pedrini, Z. Dias, A. Rocha. "Authentication of Vincent van Gogh's work," in International Conference on Computer Analysis of Images and Patterns (CAIP), 2021, pp. 371--380

paper doi

L. David. "A Study of the ISOMAP Algorithm and Its Applications in Machine Learning," in UFSCar, Universidade Federal de São Carlos, 2015

paper code

Reports

MO444 - Final Report: Classificando impacto humano sobre a floresta amazônica a partir de imagens de satélite

Application of multiple strategies involving deep convolutional networks in the solution of the "Planet: Understanding the Amazon from Space" challenge.

Published at Jun 19, 2021 Satellite Imagery Multi-label Convolutional Networks Portuguese

MO443 - Assignment 4: Texture Descriptors

Application of texture-describing methods to extract descriptors of texture samples.

Published at Jun 16, 2021 Computer Vision Texture

MO443 - Assignment 3: Image Segmentation

Segmentation of geometric shapes and blood cells using classic Computer Vision techniques.

Published at Jun 12, 2021 Computer Vision Segmentation

MO443 - Assignment 2: Discrete Fourier Transform

Application of the Discrete Fourier Transform over images using TensorFlow.

Published at May 7, 2021 Computer Vision FFT

MO443 - Assignment 1: Image Manipulation

Implementation of basic operations over images using Python programming language.

Published at Apr 6, 2021 Computer Vision Vectorization

Connoisseur: Provenance Analysis in Paintings (MSc dissertation)

A comparative analysis of multiple convolutional networks-based strategies to identify and classify art-related digital artifacts. In a first stage, we conduct our experiments over the VGDB2016 dataset, comprising art pieces from Vincent van Gogh and his peers. We then expand our strategy to a broader multi-class scenario, seeking to distinguish paintings from 1,584 different authors.

Published at Sep 23, 2019

Rede Neural Convolucional Paralelizada em PyCUDA

GPU implementations for the mathematical operations commonly employed in the development of convolutional networks.

Published at Dec 14, 2017 CUDA Portuguese

Redes Artificiais de Arquitetura Variante

Searching network architectures using Hill-Climbing and Genetic Algorithms.

Published at Dec 5, 2016 Portuguese Local Optimization Genetic Algorithms

Treinamento de redes neurais como um processo evolutivo

Exemplifying the training of neural networks as an evolutionary process. Network parameters are modeled as individuals and genetic algorithm is used to optimize the population accordingly.

Published at Jun 1, 2016 Portuguese Genetic Algorithms

A Study of the ISOMAP Algorithm and Its Applications in Machine Learning (Undergraduate final presentation)

Presentation of linear and non-linear classic dimensionality reduction methods.

Published at Dec 7, 2015 Dimensionality Reduction Manifold Learning ISOMAP

A Study of the ISOMAP Algorithm and Its Applications in Machine Learning (Undergraduate final report)

A study of the foundations of nonlinear dimensionality reduction through Manifold Learning.

Published at Dec 7, 2015 Dimensionality Reduction Manifold Learning ISOMAP

Educação e tecnologia: desafios para a construção de conhecimento por meio da computação

Será abordada a relação entre educação e computação, que se torna cada vez mais próxima. Mais precisamente, os objetos de aprendizagem, elementos essenciais nos processos de ensino e aprendizagem da educação a distância ou projetos de educação complementar.

Published at Jan 15, 2013 Portuguese Education

Events

17th International Conference on Computer Vision Theory and Applications (Feb. 2022)

"MinMax-CAM: Improving Focus of CAM-based Visualization Techniques in Multi-label Problems"

Speakerhttps://visapp.scitevents.org/?y=2022

IEEE Symposium Series on Computational Intelligence (Dec. 2021)

"Connoisseur: Provenance Analysis in Paintings," presented in the Applications of Deep Learning II track.

Speakerhttps://attend.ieee.org/ssci-2021/

The 19th International Conference on Computer Analysis of Images and Patterns (Sep. 2021)

"Authentication of Vincent van Gogh's Work," presented in the Biometrics, Cryprography and Security track.

Speakerhttps://cyprusconferences.org/caip2021/

Dextra's Data Week (Nov. 2020)

"Superpowers with Natural Language Processing," illustrating commercial cases for natural language processing techniques and algorithms.

Speakerhttps://www.dextra.com.br/

DevFest, GDG Campinas (Nov. 2018)

Participant in Computing Conference organized by GDG Campinas.

Participanthttps://gdg-campinas.github.io/devfest/

3rd Online National Congress on Mechanic Engineer and Automation (CONAENGE 2018)

Co-speaker for poster titled "Anomaly Detection and Failure Prediction in Industrial Machinery" (in Portuguese), describing the work performed at Solvian Tech for its multiple clients.

Co-Speakerhttps://conaenge.com.br/conaenge-2018/

SECOMP Unicamp 2017/2

Computing Week at Universidade Estadual de Campinas.

Organizerhttps://www.facebook.com/secompunicamp/

SECOMP UFScar 2015

Computing Week at Universidade Federal de São Carlos.

Speakerhttps://www.facebook.com/secompufscar/

Projects

wikiart

Full retriever for art and metadata in http://wikiart.org/

keras-explainable

Efficient explaining AI algorithms for Keras models

mazerunner

Final Project for Artificial Intelligence 2016-1 class at UNICAMP.

minmax-cam

Official implementation for the paper "MinMax-CAM: Improving Focus of CAM-Based Visualization Techniques in Multi-Label Problems"

jmagic

A Java environment in which we (and AI agents) can play the game Magic: The Gathering.

grapher

RESTful APIs creator based on resource schematics.

manifold-learning

Final project as student at Universidade Federal de Sao Carlos.

artificial

A basic API for artificial intelligent agents described in the book "Artificial Intelligence: A Modern Approach", by Russell and Norvig.

algorithms-in-tensorflow

Experiments with gpu support, persistent logging and stable docker env.

convolutional

Convolutional Nets implemented in pyCuda.

tsp

Exploring shortest-paths in the Traveling Salesman Problem.

the-alpha

The Alpha game made in Unity3D

connoisseur

Provenance Analysis in Paintings

resnet38d-tf

Tensorflow implementation for ResNet38 dilated, with weights

grapher-vertex

A tiny demonstration of how to build with grapher.

unicamp-mo443-image-processing

Introduction to Image Processing, 2021/1

unicamp-mo444-machine-learning

Introduction to Machine Learning, 2021/1

demosthenes-legacy

Academic management application

drf-base

Base project for REST WebAPIs

tf-experiment

Tensorflow experiment template.

bread-quest

Bread Quest game (final project for GMAP 260, Drexel University)

unicamp-ia004-neural-networks-2

Neural Networks II class at Universidade Federal de Campinas

see repositories on GitHub