SPEED: Separable Pyramidal Pooling EncodEr-Decoder for Real-Time Monocular Depth Estimation on Low-Resource Settings
Speed is a critical factor in many computer vision tasks, such as scene understanding and visual odometry, which are essential components in autonomous and robotic systems. The ability to estimate depth from a single frame is called monocular depth estimation (MDE), and it is an essential skill for many computer vision applications. However, vision transformer architectures are too deep and complex for real-time inference on low-resource platforms. This is where the Separable Pyramidal pooling E