arXiv:2511.01502v1 Announce Type: new 
Abstract: Unsupervised learning of depth and ego-motion, two fundamental 3D perception tasks, has made significant strides in recent years. However, most methods treat ego-motion as an auxiliary task, either mixing all motion types or excluding depth-independent rotational motions in supervision. Such designs limit the incorporation of strong geometric constraints, reducing reliability and robustness under diverse conditions. This study introduces a discriminative treatment of motion components, leveraging the geometric regularities of their respective rigid flows to benefit both depth and ego-motion estimation. Given consecutive video frames, network outputs first align the optical axes and imaging planes of the source and target cameras. Optical flows between frames are transformed through these alignments, and deviations are quantified to impose geometric constraints individually on each ego-motion component, enabling more targeted refinement. These alignments further reformulate the joint learning process into coaxial and coplanar forms, where depth and each translation component can be mutually derived through closed-form geometric relationships, introducing complementary constraints that improve depth robustness. DiMoDE, a general depth and ego-motion joint learning framework incorporating these designs, achieves state-of-the-art performance on multiple public datasets and a newly collected diverse real-world dataset, particularly under challenging conditions. Our source code will be publicly available at mias.group/DiMoDE upon publication.

أحرزت دراسة حديثة تقدمًا كبيرًا في التعلم غير الخاضع للإشراف للعمق وحركة الأنا، وهما أمران أساسيان للإدراك ثلاثي الأبعاد. على عكس الطرق السابقة التي كانت غالبًا ما تعتبر حركة الأنا مهمة ثانوية، تقدم هذه الدراسة نهجًا أكثر دقة يعزز موثوقية وفعالية هذه العمليات التعليمية. هذا مهم لأن تحسين كيفية إدراك الآلات للعمق والحركة يمكن أن يؤدي إلى أداء أفضل في تطبيقات متنوعة، من الروبوتات إلى الواقع الافتراضي.

Un estudio reciente ha logrado avances significativos en el aprendizaje no supervisado de la profundidad y el ego-movimiento, que son fundamentales para la percepción 3D. A diferencia de los métodos anteriores que a menudo trataban el ego-movimiento como una tarea secundaria, esta investigación introduce un enfoque más refinado que mejora la fiabilidad y robustez de estos procesos de aprendizaje. Esto es importante porque mejorar cómo las máquinas perciben la profundidad y el movimiento puede llevar a un mejor rendimiento en diversas aplicaciones, desde la robótica hasta la realidad virtual.

Une étude récente a réalisé des avancées significatives dans l'apprentissage non supervisé de la profondeur et de l'ego-mouvement, qui sont cruciaux pour la perception 3D. Contrairement aux méthodes précédentes qui considéraient souvent l'ego-mouvement comme une tâche secondaire, cette recherche propose une approche plus raffinée qui améliore la fiabilité et la robustesse de ces processus d'apprentissage. Cela est important car améliorer la façon dont les machines perçoivent la profondeur et le mouvement peut conduire à de meilleures performances dans diverses applications, de la robotique à la réalité virtuelle.

A recent study has made significant advancements in the unsupervised learning of depth and ego-motion, which are crucial for 3D perception. Unlike previous methods that often treated ego-motion as a secondary task, this research introduces a more refined approach that enhances the reliability and robustness of these learning processes. This is important because improving how machines perceive depth and motion can lead to better performance in various applications, from robotics to virtual reality.

Discriminately Treating Motion Components Evolves Joint Depth and Ego-Motion Learning

Was this article worth reading? Share it

Mapfit

GPTHumanizer

LucidQuery AI