Involution: Inverting the Inherence of Convolution for Visual Recognition

Convolution has been the core ingredient of modern neural networks, triggering the surge of deep learning in vision. In this work, we rethink the inherent principles of standard convolution for vision tasks, specifically spatial-agnostic and channel-specific. Instead, we present a novel atomic operation for deep neural networks by inverting the aforementioned design principles of convolution, coined as involution. We additionally demystify the recent popular self-attention operator and subsume it into our involution family as an over-complicated instantiation. The proposed involution operator could be leveraged as fundamental bricks to build the new generation of neural networks for visual recognition, powering different deep learning models on several prevalent benchmarks, including ImageNet classification, COCO detection and segmentation, together with Cityscapes segmentation. Our involution-based models improve the performance of convolutional baselines using ResNet-50 by up to 1.6% top-1 accuracy, 2.5% and 2.4% bounding box AP, and 4.7% mean IoU absolutely while compressing the computational cost to 66%, 65%, 72%, and 57% on the above benchmarks, respectively. Code and pre-trained models for all the tasks are available at

PDF Abstract CVPR 2021 PDF CVPR 2021 Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Classification ImageNet RedNet-38 Top 1 Accuracy 77.6% # 776
Number of params 12.4M # 476
GFLOPs 2.2 # 151
Image Classification ImageNet RedNet-26 Top 1 Accuracy 75.9% # 833
Number of params 9.2M # 442
GFLOPs 1.7 # 132
Image Classification ImageNet RedNet-101 Top 1 Accuracy 79.1% # 690
Number of params 25.6M # 574
GFLOPs 4.7 # 217
Image Classification ImageNet RedNet-50 Top 1 Accuracy 78.4% # 744
Number of params 15.5M # 491
GFLOPs 2.7 # 164
Image Classification ImageNet RedNet-152 Top 1 Accuracy 79.3% # 682
Number of params 34M # 630
GFLOPs 6.8 # 242