Olivier Canévet | EPFL Graph Search

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Multi-scale sequential network for semantic text segmentation and localization

Olivier Canévet, Jean-Marc Odobez

We present a novel method for semantic text document analysis which in addition to localizing text it labels the text in user-defined semantic categories. More precisely, it consists of a fully-convolutional and sequential network that we apply to the particular case of slide analysis to detect title, bullets and standard text. Our contributions are twofold: (1) A multi-scale network consisting of a series of stages that sequentially refine the prediction of text and semantic labels (text, title, bullet); (2) A synthetic database of slide images with text and semantic annotation that is used to train the network with abundant data and wide variability in text appearance, slide layouts, and noise such as compression artifacts. We evaluate our method on a collection of real slide images collected from multiple conferences, and show that it is able to localize text with an accuracy of 95%, and to classify titles and bullets with accuracies of 94% and 85% respectively. In addition, we show that our method is competitive on scene and born-digital image datasets, such as ICDAR 2011, where it achieves an accuracy of 91.1%.

2020

WatchNet plus plus : efficient and accurate depth-based network for detecting people attacks and intrusion

Olivier Canévet, Jean-Marc Odobez

We present an efficient and accurate people detection approach based on deep learning to detect people attacks and intrusion in video surveillance scenarios Unlike other approaches using background segmentation and pre-processing techniques, which are not able to distinguish people from other elements in the scene, we propose WatchNet++ that is a depth-based and sequential network that localizes people in top-view depth images by predicting human body joints and pairwise connections (links) such as head and shoulders. WatchNet++ comprises a set of prediction stages and up-sampling operations that progressively refine the predictions of joints and links, leading to more accurate localization results. In order to train the network with varied and abundant data, we also present a large synthetic dataset of depth images with human models that is used to pre-train the network model. Subsequently, domain adaptation to real data is done via fine-tuning using a real dataset of depth images with people performing attacks and intrusion. An extensive evaluation of the proposed approach is conducted for the detection of attacks in airlocks and the counting of people in indoors and outdoors, showing high detection scores and efficiency. The network runs at 10 and 28 FPS using CPU and GPU, respectively.

2020