Detailed schedule

The detailed schedule for the DIRA workshop at CVPR 2020, which takes place on June 14, 2020 (time slots below are all in pacific time).

Please note that all our talks are pre-recorded (1 opening remarks, 7 keynotes, 1 contributed talk and 7 poster presentations).

Please click the title provided in the schedule below for the corresponding talk video and slides. ENJOY!

For CVPR 2020 attendees, we highly recommend you watch the videos beforehand, so you will be able to make great use of the opportunities to have direct interactions with the author(s) of the talks during their scheduled time slot (you need to go back to the CVPR DIRA workshop landing page for interactions between the authors and the audience during their scheduled time slot).

Table of Contents

8:30 – 9:00 Opening remarks by Liping Yang

Why DIRA?

Overview of why diagram image retrieval and analysis (DIRA) matters and my recent novel DIRA work, as well as DIRA workshop summary at CVPR 2020.

Author: Liping Yang

Keywords:  Diagram images, Technical drawings, Image representation, Image classification, Line segment detection, Image retrieval, Visual similarity, Image analysis

 

9:00 – 9:40 Keynote 1  by Dr. Timothy Hospedales from U of Edinburgh

Free-Hand Sketch Analysis

I will introduce the area of free-hand sketch analysis including fun applications as well as classic and state of the art methodologies.

Author: Timothy Hospedales

Keywords:  Free-Hand Sketch Analysis, Sketch Recognition, Sketch-Based Image Retrieval, Sketch Generation, Sketch Segmentation, Sketch Abstraction

 

9:40 – 10:20 Contributed talk  by Zac Yu and Adriana Kovashka

We present an interactive image search method that uses GAN-synthesized images instead of textual questions to collect relative attribute feedbacks.
 Authors: Zac Yu, Adriana Kovashka
  Keywords:  Content-Based Image Retrieval,Interactive Image Search,Relative Attribute,Generative Adversarial Network,Image Synthesis,Image Editing,Relevance Feedback,Computer Vision,Human-Computer Interaction

10:20 – 11:00 Keynote 2  by Dr. Rogerio Feris from IBM

Visual Learning Beyond Natural Images 

Visual Learning Beyond Natural Images

 Author: Rogerio Feris

Keywords:  Visual Learning Beyond Natural Images, Cross-domain Few-shot Learning, Transfer Learning, Multi-Task Learning

 

11:00 – 11:20 LIVE Keynote QA1  

 

11:20 – 12:20 Poster session

Diagram Image Retrieval and Analysis: Challenges and Opportunities (by Liping Yang, Ming Gong, and Vijayan Asari)

Diagram Image Retrieval and Analysis: Challenges and Opportunities A systematic review of key recent research on diagram image retrieval and analysis, with demonstration and discussion of challenges and opportunities.

 Authors: Liping Yang, Ming Gong, Vijayan K Asari

Keywords:  Diagram images, Technical drawings, Systematic review, Shape descriptor, CBIR, Topology and geometry, Visual similarity, Patent images

Syntharch: Interactive Image Search with Attribute-Conditioned Synthesis (by Zac Yu and Adriana Kovashka)

We present an interactive image search method that uses GAN-synthesized images instead of textual questions to collect relative attribute feedbacks.

 Authors: Zac Yu, Adriana Kovashka

Keywords:  Content-Based Image Retrieval,Interactive Image Search,Relative Attribute,Generative Adversarial Network,Image Synthesis,Image Editing,Relevance Feedback,Computer Vision,Human-Computer Interaction

Learning Spatial Relationships between Samples of Image Shapes (by Juan Castorena, Manish Bhattarai, and Diane Oyen)

This work presents a patent document classification/retrieval method based on image data by learning geometric shape relationships through graph CNN.

Authors: Juan Castorena, Manish Bhattarai, Diane Oyen

Keywords:  Computer vision, Machine Learning, Classification, Retrieval, Graph Neural Networks, Patent Images.

Diagram Image Retrieval using Sketch-Based Deep Learning and Transfer Learning (by Manish Bhattarai, Diane Oyen, Juan Castorena, Liping Yang, and Brendt Wohlberg)

TSNE projection of image at various stages i) Input image space, ii) Encoder output feature space and iii) Siamese tuned output feature space.

Authors: Manish Bhattarai, Diane Oyen, Juan Castorena, Liping Yang, Brendt Wohlberg

Keywords:  Diagram image retrieval, Zero-shot/One-shot learning, transfer learning, domain generalization, patent images, Scientific drawings, deep learning.

Automatic Digitization of Engineering Diagrams using Deep Learning and Graph Search (by Shouvik Mani, Michael Haddad, Dan Constantini, Willy Douhard, Qiwei Li , and Louis Poirier)

A computer vision pipeline for automatically digitizing Piping and Instrumentation Diagrams (P&IDs).

Authors: Shouvik Mani, Michael Haddad, Dan Constantini, Willy Douhard, Qiwei Li, Louis Poirier

Keywords:  engineering diagram, P&ID, deep learning, CNN, symbol detection, graph search, text recognition

Structured Query-Based Image Retrieval Using Scene Graphs (by Brigit Schroeder and Subarna Tripathi)

We present a method which uses scene graph embeddings as the basis for image retrieval where visual relationships are used as structured queries.

Authors: Brigit Schroeder, Subarna Tripathi

Keywords:  scene graph, visual relationship, image retrieval, graph convolutional neural network, scene graph embedding

A Simplified Framework for Zero-shot Cross-Modal Sketch Data Retrieval (by Ushasi Chaudhuri, Biplab Banerjee, Avik Bhattacharya, and Datcu Mihai)

We tackle the problem of zero-shot cross-modal retrieval involving color and sketch images through a novel deep representation learning technique.

Authors: Ushasi Chaudhuri, Biplab Banerjee, Avik Bhattacharya, Datcu Mihai

Keywords:  Sketch, neural networks, Sketch-based image retrieval, cross-modal retrieval, Deep-learning

13:20 – 14:00 Keynote 3  by Dr. Lingfei Wu from IBM

Deep Learning on Graphs in Natural Language Processing and Computer Vision

In this talk, I described the basic idea of deep learning on graphs and introduced how to use them for question generation and grounded video description

 Author: Lingfei Wu

Keywords:  Deep Learning on Graphs, Graph Neural Networks, Natural Language Processing, Computer Vision, Question Generation, Grounded Video Description

14:00 – 14:45 Keynote 4  by Prof.  Adriana Kovashka from U of Pittsburgh

Reasoning about Complex Media from Weak Multi modal Supervision

We propose to better understand creative ads with hidden messages over SOTA methods through weak multimodal supervision and common sense reasoning.

Author: Adriana Kovashka

Keywords:  visual reasoning, vision and language, common sense, visual persuasion, weak supervision

14:45 – 15:45 Keynote 5  by Prof. Devi Parikh from Georgia Tech

AI Systems That Can See And Talk

Overview of why problems at the intersection of vision and language are exciting, what capabilities today’s AI systems have, and what challenges remain

Author: Devi Parikh

Keywords:  Vision and language, Visual Question Answering, VQA, transformer, BERT, referring expressions, visual dialog, image captioning, demo

 

15:45 – 16:15 Keynote 6  by Ranjay Krishna from Stanford University,

Scene Graphs as a Symbolic Visual Representation

Representing the visual world with scene graphs – a compositional, interpretable, structured knowledge representation.

 Author: Ranjay Krishna

Keywords:  scene graphs, compositionality, interpretability, knowledge representations, vision and language, image retrieval, action recognition, few-shot learning, visual question answering

 

16:15 – 17:00 Keynote 7  by Prof. William T. Freeman from MIT

Generating and interpreting line drawings across variations in style: two case studies

Two case studies of interpreting and generating line drawings: occlusion-aware, example-based shape interpretation, and line drawing style translation

 Author: William T. Freeman

Keywords:  line drawings, example-based methods, shape interpretation, style and content, belief propagation

17:00 – 17:20 LIVE Keynote QA2

11:00 – 11:20 LIVE Poster QA