This page summarizes all the activities we have developed in the context of CAPES STIC-AmSud project (2019-2020).

Optimized Deep Learning based Representations for Computer Vision Problems (DLRCV)

Meeting 2

  • Place: Rouen, France
  • Date: December 9-19, 2019
  • Attendees:
  • Activities:

    • Monday, December 9th, 2019

    • (Morning): Visit of the University of Rouen and the Litis Lab.

      Picture at Litis Lab (Rouen University) from left to right and top to bottom: Prof. José Saavedra, Prof. Alceu de Souza Britto, Prof. Laurent Heutte, Prof. Iván Sipirán Mendoza, Prof. Violeta Chang, Prof. Caroline Petitjean, and Prof. Analí Alfaro Alfaro

    • (Afternoon): Working meeting on Pattern Spotting to finish the paper "Two Deep Approaches for Image Retrieval and Pattern Spotting" to be submitted to the Expert Systems with Applications.

      Talk by Violeta Chang on deep learning and segmentation of medical images and working meeting.

      • Talk of Prof. Violeta Chang about Segmentation of Medical Images

    Tuesday, December 10th, 2019

    • (Morning): Meeting on Local Research Progress considering the Labs in Brazil, Chile, Peru and France.
    • (Afternoon): Presentation of Ignacio’s work submitted to PRL.

      Working meeting on the work of Alceu related to "Memory integrity of CNNs considering different fine tuning strategies".

      • Talk of Prof. Alceu Britto about memory integrity of CNNs

    Wednesday, December 11th, 2019

    • (Morning): Talk of Ivan Sipiran on deep learning and 3D and working meeting on the same topic.

      Talk of Anali Alfaro on pose estimation in videos and working meeting.

      • Talk of Prof. Anali Alfaro about pose estimation in videos

    • (Afternoon): Talk of Zacarias (PhD student in Brazil) on “deep learning for image retrieval and pattern spotting” and working meeting with Stéphane Nicolas (URN), Pierrick Tranouez (URN), Zacarias Curi Filho and all the other participants of STIC-AmSud project.

      • Talk of Zacarias Curi (Brazilian PhD Student) about image retrieval and pattern spotting

    Thursday, December 12th, 2019

    • (Morning): Talk of Laurent Heutte on dissimilarity based random forests.

      Working meeting on video summarisation and sparse coding and self-supervised learning.

      • Talk of Prof. Laurent Heutte on dissimilarity based random forests

    • (Afternoon): Seminars of the participants of the STIC-AmSud project towards LITIS lab:

      • Speaker: Prof Alceu de Souza Britto Jr, PUCPR, Curitiba, Brazil. Title: Deep Learning Approaches for Image Retrieval and Pattern Spotting in Ancient Documents. Abstract: In this work we evaluate two approaches for content-based image retrieval and pattern spotting in document images using deep learning. The first approach uses a pre-trained CNN model to cope with the lack of training data, which is fine-tuned to achieve a compact yet discriminant representation of queries and image candidates. The second approach uses a Siamese Convolution Neural Network trained on a previously prepared subset of image pairs from the ImageNet dataset to provide the similarity-based feature maps. In both methods, the learned representation scheme considers feature maps of different sizes which are evaluated in terms of retrieval performance. A robust experimental protocol using two public datasets (Tobacoo-800 and DocExplore) has shown that the proposed methods compare favorably against state-of-the-art document image retrieval and pattern spotting methods.

      Talk of Prof. Alceu Britto on deep learning for image retrieval and pattern spotting

      Speaker: José M. Saavedra, PhD, Chief R&D Officer at Orand S.A., Santiago, Chile. Title: Sketch-QNet: A Quadruplet CNN for Color Sketch Based Image Retrieval. Abstract: Sketch based image retrieval (SBIR) has been attracting the interest of many research because it represents a powerful and very natural modality for searching. However, so far the sketches were thought as simple drawings composed only by strokes, without taking into account color; nevertheless, in applications like searching in e-commerce, color is a critical attribute. Therefore, in this work we present a system for color SBIR consisting on a 3-stage training architecture with a quadruplet neural network (Sketch-QNet) devoted to learn a feature embedding that put together photos and sketches sharing color and shape. We achieve superior results compared with a method that combine deep features for shape representations with color-based descriptors. Our proposal shows a recall ratio of 0.55 compared with 0.07 achieved by the baseline, when we retrieve the first 10 photos. We also measure the mean reciprocal rank, where our proposal achieves a value of 0.3519 overcoming the baseline by 63%.

      Talk of Prof. José M. Saavedra on color sketch based image retrieval

      Speaker: Prof. Analí Alfaro and Prof. Ivan Sipiran, PUCP, Lima, Peru. Title: Similarity-based Structured Sparse Coding for Video Summarization. Abstract: We assume that a video can be seen as a subspace formed by a selected subset of its own frames. Then, each frame could be represented as a sparse linear combination of these selected frames. Our method selects those frames which contribute to the reconstruction of the entire video by taking advantage of two aspects: i) Structure and ii) Similarity between sparse codes. Structure is given by groups of frames between subtle or significant changes in the video. Similarity between sparse codes is enforced to guarantee a balanced contribution of the frames in the groups. We use these insights to propose an optimization problem which produces a sparse representation that captures the relevance of each frame in the video.

      Talk of Prof. Analí Alfaro and Prof. Ivan Sipiran on video summarization

    Friday, December 13th, 2019

    • (Morning): Talk of Caroline Petitjean on “Explainable AI for medical image segmentation and classification”.

      Talk of Ivan Sipiran on “Self-supervised image segmentation”, discussion about prior-based functional loss to regularize image segmentation. Meeting on the Zacarias Phd work. We define new ideas for instance search using deep learning based on heat-maps.

    • (Afternoon): Final meeting with all participants: project website, report, budget, planning of the next year (1 workshop in Rouen in July 2020 / 1 workshop in Recife in December 2020)

      • Meeting with all participants to plan the next year

    Monday, December 16th, 2019

    • (Morning): Working meeting on the work of Ignacio Úbeda “FPN for Document Spotting”. We discussed about the answers to reviewer of the Pattern Recognition Letters paper.

      • Working meeting on explainable AI with Violeta and Jing VANG, PhD student of Caroline.

    • (Afternoon): Working meeting on pattern spotting and deep learning with José Saavedra and Laurent Heutte.

      Working meeting with Caroline and Violeta on breast cancer detection and explainable DL.

    Tuesday, December 17th, 2019

    • (Morning): Working meeting with Zacarias Curi Filho on pattern spotting (José, Laurent).

      Talk of Jing Vang on explainable AI for image segmentation (Caroline, Violeta).

    • (Afternoon): Skype meeting with Violetas’s student about sperm cell segmentation (Caroline, Violeta, Laurent, José).

    Wednesday, December 18th, 2019

    • (Morning): Working meeting with Caroline and Violeta on HER2 breast cancer biopsies classification.

      Working meeting on sketch based image retrieval (José, Laurent).

      • Meeting of Prof. Violeta and Prof. Caroline

    • (Afternoon): Working meeting about review of experimental framework of Violeta’s student thesis on DL for medical image segmentation (Caroline, Laurent, José, Violeta).

    Thursday, December 19th, 2019

    • (Morning): Working meeting to review the final version of the Ignacio’s PRL paper (Laurent, José).

      • Working meeting on explainable AI for image segmentation (Caroline, Violeta).

  • Go to Main Page ...