Optimize Vision Transformers Architecture via Efficient Attention Modules: A Study on the Monocular Depth Estimation Task

🔋 Two modifications of METER attention module are proposed and implemented to build two more efficient versions: Meta-METER and Pyra-METER.