Exploring Data Aggregation and Transformations to Generalize across Visual Domains
This addresses the problem of poor model performance when train and test distributions differ, which is critical for deploying computer vision systems in real-world, varied environments, though it appears incremental within existing research topics.
The paper tackles dataset bias in computer vision by proposing new frameworks for Domain Generalization and Domain Adaptation using feature aggregation and visual transformations, and demonstrates that these solutions outperform state-of-the-art approaches in established benchmarks.
Computer vision has flourished in recent years thanks to Deep Learning advancements, fast and scalable hardware solutions and large availability of structured image data. Convolutional Neural Networks trained on supervised tasks with backpropagation learn to extract meaningful representations from raw pixels automatically, and surpass shallow methods in image understanding. Though convenient, data-driven feature learning is prone to dataset bias: a network learns its parameters from training signals alone, and will usually perform poorly if train and test distribution differ. To alleviate this problem, research on Domain Generalization (DG), Domain Adaptation (DA) and their variations is increasing. This thesis contributes to these research topics by presenting novel and effective ways to solve the dataset bias problem in its various settings. We propose new frameworks for Domain Generalization and Domain Adaptation which make use of feature aggregation strategies and visual transformations via data-augmentation and multi-task integration of self-supervision. We also design an algorithm that adapts an object detection model to any out of distribution sample at test time. With through experimentation, we show how our proposed solutions outperform competitive state-of-the-art approaches in established DG and DA benchmarks.