Deep Learning to See: Towards New Foundations of Computer Vision
This work addresses foundational issues in computer vision for researchers, proposing a shift from application-focused methods to theoretical frameworks, but it is incremental as it builds on existing critiques without presenting new empirical results.
The authors critique the current reliance on deep learning and labeled data in computer vision, arguing that true scientific progress requires investigating vision through information-based natural laws and developing learning theories that account for the spatiotemporal nature of visual signals.
The remarkable progress in computer vision over the last few years is, by and large, attributed to deep learning, fueled by the availability of huge sets of labeled data, and paired with the explosive growth of the GPU paradigm. While subscribing to this view, this book criticizes the supposed scientific progress in the field and proposes the investigation of vision within the framework of information-based laws of nature. Specifically, the present work poses fundamental questions about vision that remain far from understood, leading the reader on a journey populated by novel challenges resonating with the foundations of machine learning. The central thesis is that for a deeper understanding of visual computational processes, it is necessary to look beyond the applications of general purpose machine learning algorithms and focus instead on appropriate learning theories that take into account the spatiotemporal nature of the visual signal.