Stereo matching has been a fundamental topic in computer vision for almost half a century: computing a dense disparity map from two rectified images. It plays a key role in many applications, including autonomous driving, robotics, and augmented reality.
Existing studies categorize end-to-end architectures into 2D and 3D classes according to cost-volume calculation and optimization methodologies. These studies also highlight questions that remain to be answered, providing important insights into this rapid change. Stimulated by innovations in other areas of deep learning, new approaches and paradigms have emerged in the field, and the area has seen tremendous growth since then. Examples of the field's evolution, such as iterative refinement and transformer-based architectures, that show the potential for further improvements in accuracy and efficiency instill a sense of optimism and hope for the future of deep stereo matching. As deep stereo matching has progressed, many issues have surfaced, despite the impressive achievements. Lack of generalization, especially when dealing with domain transitions between real and synthetic data, is a major issue mentioned in previous studies.
Previous research conducted in the late 2010s covered the early stages of this revolution, but the following five years of research have seen even more revolutionary advances in the field. A new study from a team at the University of Bologna, a leading group in the field, shows that:
- An in-depth analysis of recent advances in deep stereo matching, with particular attention to revolutionary paradigm shifts that have changed the game in the 2020s, including the use of transformer-based architectures and groundbreaking architectural designs like RAFT-new stereo.
- We will analyze the main problems that have arisen from these advancements, break them all down, and consider the best ways to solve them.
The main findings of their paper are highlighted as follows:
Architectural Design: The benchmark results show that the RAFT-new stereo design approach is innovative and significantly improves resilience to domain changes. The team expects more frameworks to follow this new paradigm, as most of the latest frameworks published a few months before this study used it. However, the search for innovative and efficient designs is a fascinating journey that continues to engage the field, as shown by the latest proposals that always yield improved results.
RGB Enhanced Audio: The novel concept of utilizing thermal, multispectral, or event camera images as input to stereo matching networks has gained popularity over the past five years, injecting new ideas into an established but dynamic field. While this trend is encouraging, the online world requires further refinement for these new tasks.
Some of the problems predicted in previous work still exist, despite numerous successes in addressing them. The Booster dataset showed how processing high-resolution images remains challenging and how non-Lambertian objects are significant, mainly due to the lack of training data and better methods to address them. Similarly, harsh weather conditions can still be a problem.
The team notes that despite developing underlying visual models for other computer vision tasks, stereo matching is still needed: while there has been no work on stereo in this area yet, there has been some work on single-image depth estimation.
By highlighting the most effective methods currently in use, this study not only illuminates existing obstacles, but also suggests promising avenues for further research. Novices and seasoned professionals alike will find useful information and inspiring ideas in this survey, and the team hopes it will ignite a passion for pushing the boundaries of deep stereo matching.
Please check paper. All credit for this research goes to the researchers of this project. Also, don't forget to follow us. twitter.
participate Telegram Channel and LinkedIn GroupsUp.
If you like our work, you will love our Newsletter..
Please join us 46k+ ML Subreddit
Dhanshree Shenwai is a Computer Science Engineer with extensive experience in FinTech companies covering the domains of Finance, Cards & Payments, Banking and has a keen interest in the applications of AI. She is passionate about exploring new technologies and advancements in today's evolving world that will make life easier for everyone.
