Lecture

Document Segmentation: dhSegment

Graph Chat

Chat with Graph Search

Ask any question about EPFL courses, lectures, exercises, research, news, etc. or try the example questions below.

DISCLAIMER: The Graph Chatbot is not programmed to provide explicit or categorical answers to your questions. Rather, it transforms your questions into API requests that are distributed across the various IT services officially administered by EPFL. Its purpose is solely to collect and recommend relevant references to content that you can explore to help you answer your questions.

In course

HUM-369: Digital humanities

Ce cours explore comment les Humanités se transforment à l'ère de l'intelligence artificielle et des données massives. Chaque séance aborde le futur d'une discipline. Les étudiants développent un prototype numérique (Vibe coding) pour accelerer la recherch ...

Description

This lecture covers the development of dhSegment, an open-source package for document segmentation using PyTorch. Originally implemented in TensorFlow, dhSegment was ported to PyTorch to integrate state-of-the-art architectures more easily. The lecture discusses the challenges in automatic document analysis, the transition from TensorFlow to PyTorch, and the improvements made to dhSegment. Specific topics include image semantic segmentation, object recognition, text recognition, and analysis systems.

Instructors (3)

Frédéric Kaplan

Expertise Digital Humanities, Big Data, Models of language acquisition and evolution, Innovative Interfaces, Artificial Intelligence, Robotics, Developmental Systems, Intrinsic Motivation, Multi-agent system, Joint attention, Gesture Interaction Expertise Digital Humanities, Big Data, Models of language acquisition and evolution, Innovative Interfaces, Artificial Intelligence, Robotics, Developmental Systems, Intrinsic Motivation, Multi-agent system, Joint attention, Gesture Interaction Short biography Pro

Isabella Di Lenardo

Ph.D. in Theories and Art History. Isabella di Lenardo is senior scientist in Digital Humanities and Urban History. Her research activity is focused on digital tools and methods applied to Urban History. She is an expert in ancient cartography, city representations, cadastral sources interpreted through digital modeling, extraction and analysis systems.She's also interested in network analysis questioning the production and circulation of artistic and architectural knowledge in Europe XVIth – XVIIIth centuries in particular on North-South relationships and influences. The development of collaboration projects with European institutions and the activity carried out on various initiatives supported by the European Commission have allowed her to acquire specialism in the field of science applied to heritage, developing in particular specific skills for the enhancement of Cultural Heritage through digital tools. She leads projects in collaboration with Bibliothèque national de France (Paris), Institut National d’Histoire de l’Art (Paris), Ecole nationale des chartes (Paris), Université Paris I Panthéon-Sorbonne, Centre Allemand d’Histoire de l’Art (Paris), Musée du Louvre, Archives Nationales de Paris, Bibliothèque Historique de la Ville de Paris, Réunion des Musées Nationaux. She supervised all the urban modeling simulations for the Venice Time Machine project and acted as content curator for all the exhibitions (Venice Biennale, Grand Palais-Paris, Datasquare - ArtLab-Lausanne, Museo Correr-Venice). She was Head of “Replica” project : digitizing 1 million photos of artworks in the Fondazione Giorgio Cini (Venice) and visual patterns extraction through a search engine for visual similarities. She has a Ph.D in Art History and also studied Archaeology and Urban Studies. She has published essays and articles about Venetian Art and Urban History and participated in many Art exhibitions in European museums. She coordinated summer schools and workshops about Digital Tools for Art History and Urban History. She attended conferences on digital methods for History in universities and patrimonial institutions.

Official source

Related concepts (32)

Image analysis

Image analysis or imagery analysis is the extraction of meaningful information from s; mainly from s by means of techniques. Image analysis tasks can be as simple as reading bar coded tags or as sophisticated as identifying a person from their face. Computers are indispensable for the analysis of large amounts of data, for tasks that require complex computation, or for the extraction of quantitative information.

TensorFlow

TensorFlow is a free and open-source software library for machine learning and artificial intelligence. It can be used across a range of tasks but has a particular focus on training and inference of deep neural networks. TensorFlow was developed by the Google Brain team for internal Google use in research and production. The initial version was released under the Apache License 2.0 in 2015. Google released the updated version of TensorFlow, named TensorFlow 2.0, in September 2019.

Image segmentation

In and computer vision, image segmentation is the process of partitioning a into multiple image segments, also known as image regions or image objects (sets of pixels). The goal of segmentation is to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze. Image segmentation is typically used to locate objects and boundaries (lines, curves, etc.) in images. More precisely, image segmentation is the process of assigning a label to every pixel in an image such that pixels with the same label share certain characteristics.

Market segmentation

In marketing, market segmentation is the process of dividing a broad consumer or business market, normally consisting of existing and potential customers, into sub-groups of consumers (known as segments) based on shared characteristics. In dividing or segmenting markets, researchers typically look for common characteristics such as shared needs, common interests, similar lifestyles, or even similar s.

Deep learning

Deep learning is part of a broader family of machine learning methods, which is based on artificial neural networks with representation learning. The adjective "deep" in deep learning refers to the use of multiple layers in the network. Methods used can be either supervised, semi-supervised or unsupervised.

Related lectures (64)

Quantum and Nanocomputing

Covers quantum wire analysis, transmission function, and tunneling in dots.

Deep Learning: Convolutional Neural Networks

Covers Convolutional Neural Networks, standard architectures, training techniques, and adversarial examples in deep learning.

Data Abstractions in AEC: Connecting with Speckle

By the instructor Claire Kuang explores data abstractions in the AEC industry and the role of Speckle in connecting industry players through an open-source platform.

Open Source in Switzerland: Impact and Examples

Explores the impact of open source practices in Switzerland, focusing on their relevance in practical applications and academia.

Open-Source Architecture Community: BlenderBIM

Delves into the open-source architecture community, focusing on BlenderBIM and various tools used in the architectural field.

Maud Ehrmann

Maud Ehrmann is a research scientist at EPFL’s Digital Humanities Laboratory lead by Prof. Frédéric Kaplan . She holds a PhD in Computational Linguistics from the Paris Diderot Universtiy (Paris 7) and has been engaged in a large number of scientific projects centred on information extraction and text analysis, both for present-time and historical documents. Her main research interests span Natural Language Processing and Digital Humanities and include, among others, historical text annotation, historical data processing and representation, named entity recognition, and multilingual linguistic resources creation. Her current work at the DHLAB focuses on ‘impresso - Media Monitoring of the Past’ , a SNF sinergia project she initiated with Marten Düring ( C2DH ) and Simon Clematide ( ICL ) and which aims at enabling critical analysis of historical newspapers. In addition to the overall project management, her contributions to this project include system design and data management, annotation and benchmarking and named entity processing. Besides, she participates to the activities of the Venice Time Machine , working particularly on information extraction and knowledge representation tasks. Previously, she worked on the Garzoni project where she supervised and contributed to the development of a web-based transcription and annotation interface - in collaboration with Orlin Topalov, and built a linked data-based historical knowledge base. She also contributed to the Le Temps Digital Archives project . Prior to joining the DHLAB, she worked at the Linguistics Computing Laboratory at the Sapienza University of Rome with Roberto Navigli, where she worked on the BabelNet resource - a very large multilingual encyclopaedic dictionary and semantic network - and contributed to the LIDER project. Before that, she has been working for four years at the European Commission’s Joint Research Centre in Ispra, Italy, as member of the OPTIMA unit (now Text and Data mining unit), which develops innovative and application-oriented solutions for retrieving and extracting information from the Internet with a focus on high multilinguality. Together with Erik van der Goot, Ralf Steinberger , Hristo Tanev, Leo della Rocca and many others, she contributed to the development of the Europe Media Monitor (EMM). Prior, she worked at the Xerox Europe Research Centre in Grenoble, France (now Naver Labs Europe ) in the Parsing & Semantics group led by Frédérique Segond, first as PhD candidate supported through a CIFRE grant under the supervision of Caroline Brun and Bernard Victorri , then as a post-doctoral researcher. There her research focused mainly on the automatic processing and fine-grained analysis of entities of interest, specifically named entities and temporal expressions.

Related concepts (32)