Involution: Inverting the Inherence of Convolution for Visual Recognition ( Research Paper Explained)
, involution, computervision, attention Convolutional Neural Networks (CNNs) have dominated computer vision for almost a decade by applying two fundamental principles: Spatial agnosticism and channelspecific computations. Involution aims to invert these principles and presents a spatialspecific computation, which is also channelagnostic. The resulting Involution Operator and RedNet architecture are a compromise between classic Convolutions and the newer Local SelfAttention architectures and perform favorably in terms of computation accuracy tradeoff when compared to either. OUTLINE: 0:00 Intro Overview 3:00 Principles of Convolution 10:50 Towards spatialspecific computations 17:00 The Involution Operator 20:00 Comparison to SelfAttention 25:15 Experimental Results 30:30 Comments Conclusion Paper: Code: Abstract: Convolution has been the core ingredient of
|
|