# FMDConv: Fast Multi-Attention Dynamic Convolution via Speed-Accuracy Trade-off

arXiv: https://arxiv.org/abs/2503.17530 · HTML: https://mlsanity.com/paper/2503.17530/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 20 of its 47 references that are in the map.
- First posted on arXiv 21 Mar 2025.

## Builds on (most cited first, 20 of 20)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [Squeeze-and-Excitation Networks](https://mlsanity.com/paper/1709.01507/) (2017)
- [Going Deeper with Convolutions](https://mlsanity.com/paper/1409.4842/) (2014)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices](https://mlsanity.com/paper/1707.01083/) (2017)
- [RepVGG: Making VGG-style ConvNets Great Again](https://mlsanity.com/paper/2101.03697/) (2021)
- [Dynamic Neural Networks: A Survey](https://mlsanity.com/paper/2102.04906/) (2021)
- [Dynamic Convolution: Attention over Convolution Kernels](https://mlsanity.com/paper/1912.03458/) (2019)
- [CondConv: Conditionally Parameterized Convolutions for Efficient Inference](https://mlsanity.com/paper/1904.04971/) (2019)
- [cosFormer: Rethinking Softmax in Attention](https://mlsanity.com/paper/2202.08791/) (2022)
- [Speed/accuracy trade-offs for modern convolutional object detectors](https://mlsanity.com/paper/1611.10012/) (2016)
- [Global Second-order Pooling Convolutional Networks](https://mlsanity.com/paper/1811.12006/) (2018)
- [Omni-Dimensional Dynamic Convolution](https://mlsanity.com/paper/2209.07947/) (2022)
- [Partial Order Pruning: for Best Speed/Accuracy Trade-off in Neural Architecture Search](https://mlsanity.com/paper/1903.03777/) (2019)
- [Revisiting Dynamic Convolution via Matrix Decomposition](https://mlsanity.com/paper/2103.08756/) (2021)
- [Laser: Efficient Language-Guided Segmentation in Neural Radiance Fields](https://mlsanity.com/paper/2501.19084/) (2025)
- [MASK-CNN-Transformer For Real-Time Multi-Label Weather Recognition](https://mlsanity.com/paper/2304.14857/) (2023)
- [Sentinel-Guided Zero-Shot Learning: A Collaborative Paradigm without Real Data Exposure](https://mlsanity.com/paper/2403.09363/) (2024)
- [Activation Function Optimization Scheme for Image Classification](https://mlsanity.com/paper/2409.04915/) (2024)

## Built on by (most cited first, 1 of 1)

- [TRACE: Temporally Reliable Anatomically-Conditioned 3D CT Generation with Enhanced Efficiency](https://mlsanity.com/paper/2507.00802/) (2025)

## Abstract (arXiv)

Spatial convolution is fundamental in constructing deep Convolutional Neural Networks (CNNs) for visual recognition. While dynamic convolution enhances model accuracy by adaptively combining static kernels, it incurs significant computational overhead, limiting its deployment in resource-constrained environments such as federated edge computing. To address this, we propose Fast Multi-Attention Dynamic Convolution (FMDConv), which integrates input attention, temperature-degraded kernel attention, and output attention to optimize the speed-accuracy trade-off. FMDConv achieves a better balance between accuracy and efficiency by selectively enhancing feature extraction with lower complexity. Furthermore, we introduce two novel quantitative metrics, the Inverse Efficiency Score and Rate-Correct Score, to systematically evaluate this trade-off. Extensive experiments on CIFAR-10, CIFAR-100, and ImageNet demonstrate that FMDConv reduces the computational cost by up to 49.8% on ResNet-18 and 42.2% on ResNet-50 compared to prior multi-attention dynamic convolution methods while maintaining competitive accuracy. These advantages make FMDConv highly suitable for real-world, resource-constrained applications.