One Day Meeting: Synthetic Data for Machine Learning
Wednesday 8 November 2023
Chairs: Abdulrahman Kerim - UCA, Leandro Soriano Marcolino – Lancaster University, Erickson Nascimento - Universidade Federal de Minas Gerais & Microsoft
Keynote Speakers
Dr. Tadas Baltrusaitis (Principal Scientist, Microsoft) Prof. Edmond Prakash (Research Centre for Creative Technologies, UCA)
Programme
The recent success of Machine Learning (ML) models is associated with the ability to train deep models on large-scale training data. However, annotating large-scale datasets is still the bottleneck in Computer Vision (CV) and Natural Language Processing (NLP) fields. Furthermore, ensuring diverse training data under challenging attributes like adverse weather conditions or when data is scarce is not only dangerous, time-consuming, and hard to collect but also cumbersome and subjective to human errors in the annotation process. Synthetic data comes as a solution to help solve all the above issues. While its application in CV and NLP has attracted more attention, especially with the recent paradigm shift from model-centric ML to data-centric ML solutions. This Symposium explores all stages of synthetic data from its creation to use in training and as applications.
The Programme for the Meeting can be found here.
09:30 Registration and Coffee 10.00 Keynote Talk - Dr. Tadas Baltrusaitis (Principal Scientist, Microsoft)
10.45 Divyanshu Mishra (University of Oxford) - Dual Conditioned Diffusion Models for Out-of-Distribution Detection
11.15 Coffee Break
11.45 Bram Vanherle (Hasselt University) - Analysis of Training Object Detection Models with Synthetic Data
12.15 Jing Wang (Sheffield Hallam University) - Photorealistic Synthetic Image Dataset for Fall Detection
12.45 Lunch Break
13.45 Zien Ma (Cardiff University) - Quantification of Metabolites in Magnetic Resonance Spectra with Deep Learning: Insights on Simulated and Real Data
14.15 Felipe Cadar Chamone (Universidade Federal de Minas Gerais) - A New Dataset for Evaluating Keypoint Correspondence Methods on Complex and Non-rigid Scenes
14.45 Coffee Break
15.15 Keynote Talk - Prof. Edmond Prakash (Research Centre for Creative Technologies, UCA)
16.00 Benoit Quentin Arthur Vallade (Amazon Prime Video) - Closing the Domain Gap Between Development and Production Environments
16.30 Closing thoughts and final discussion
17.00 Meeting Ends
Meeting Location
The meeting will take place at:
British Computer Society (BCS), 25 Copthall Avenue, London EC2R 7BP
Registration
We keep the cost of attending these events as low as possible to ensure no barriers from the whole computer vision community attending. The registration costs are as follows
-
BMVA Members: £20
-
Non BMVA Members £40 (Includes membership to the BMVA)
Both Include Lunch and refreshments for the day