Are Large Kernels Better Teachers than Transformers for ConvNets?

• URL: http://arxiv.org/abs/2305.19412v1
• Date: Tue, 30 May 2023 21:05:23 GMT
• Title: Are Large Kernels Better Teachers than Transformers for ConvNets?
• Authors: Tianjin Huang, Lu Yin, Zhenyu Zhang, Li Shen, Meng Fang, Mykola Pechenizkiy, Zhangyang Wang and Shiwei Liu
• Abstract summary: This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small- Kernel ConvNets.
• Score: 82.4742785108714
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: This paper reveals a new appeal of the recently emerged large-kernel Convolutional Neural Networks (ConvNets): as the teacher in Knowledge Distillation (KD) for small-kernel ConvNets. While Transformers have led state-of-the-art (SOTA) performance in various fields with ever-larger models and labeled data, small-kernel ConvNets are considered more suitable for resource-limited applications due to the efficient convolution operation and compact weight sharing. KD is widely used to boost the performance of small-kernel ConvNets. However, previous research shows that it is not quite effective to distill knowledge (e.g., global information) from Transformers to small-kernel ConvNets, presumably due to their disparate architectures. We hereby carry out a first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel ConvNets, due to more similar architectures. Our findings are backed up by extensive experiments on both logit-level and feature-level KD ``out of the box", with no dedicated architectural nor training recipe modifications. Notably, we obtain the \textbf{best-ever pure ConvNet} under 30M parameters with \textbf{83.1\%} top-1 accuracy on ImageNet, outperforming current SOTA methods including ConvNeXt V2 and Swin V2. We also find that beneficial characteristics of large-kernel ConvNets, e.g., larger effective receptive fields, can be seamlessly transferred to students through this large-to-small kernel distillation. Code is available at: \url{https://github.com/VITA-Group/SLaK}.

Related papers

• Scaling Up Your Kernels: Large Kernel Design in ConvNets towards Universal Representations [17.41381592056492]
  This paper proposes the paradigm of large convolutional kernels in designing modern Convolutional Neural Networks (ConvNets) We establish that employing a few large kernels, instead of stacking multiple smaller ones, can be a superior design strategy. We propose the UniRepLKNet architecture, which offers systematical architecture design principles specifically crafted for large- Kernel ConvNets.
  arXiv  Detail & Related papers  (2024-10-10T15:43:55Z)
• Designing Concise ConvNets with Columnar Stages [33.248031676529635]
  We introduce a refreshing ConvNet macro design called Columnar Stage Network (CoSNet) CoSNet has a systematically developed simple and concise structure, smaller depth, low parameter count, low FLOPs, and attention-less operations. Our evaluations show that CoSNet rivals many renowned ConvNets and Transformer designs under resource-constrained scenarios.
  arXiv  Detail & Related papers  (2024-10-05T09:03:42Z)
• UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio, Video, Point Cloud, Time-Series and Image Recognition [61.01408259741114]
  We propose four architectural guidelines for designing large- Kernel-based convolutional neural networks (ConvNets) Our proposed large- Kernel-based ConvNet shows leading performance in image recognition. We discover large kernels are the key to unlocking the exceptional performance of ConvNets in domains where they were originally not proficient.
  arXiv  Detail & Related papers  (2023-11-27T07:48:50Z)
• Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition [158.15602882426379]
  This paper does not attempt to design a state-of-the-art method for visual recognition but investigates a more efficient way to make use of convolutions to encode spatial features. By comparing the design principles of the recent convolutional neural networks ConvNets) and Vision Transformers, we propose to simplify the self-attention by leveraging a convolutional modulation operation.
  arXiv  Detail & Related papers  (2022-11-22T01:39:45Z)
• MogaNet: Multi-order Gated Aggregation Network [64.16774341908365]
  We propose a new family of modern ConvNets, dubbed MogaNet, for discriminative visual representation learning. MogaNet encapsulates conceptually simple yet effective convolutions and gated aggregation into a compact module. MogaNet exhibits great scalability, impressive efficiency of parameters, and competitive performance compared to state-of-the-art ViTs and ConvNets on ImageNet.
  arXiv  Detail & Related papers  (2022-11-07T04:31:17Z)
• Fast-ParC: Capturing Position Aware Global Feature for ConvNets and ViTs [35.39701561076837]
  We propose a new basic neural network operator named position-aware circular convolution (ParC) and its accelerated version Fast-ParC. Our Fast-ParC further reduces the O(n2) time complexity of ParC to O(n log n) using Fast Fourier Transform. Experiment results show that our ParC op can effectively enlarge the receptive field of traditional ConvNets.
  arXiv  Detail & Related papers  (2022-10-08T13:14:02Z)
• More ConvNets in the 2020s: Scaling up Kernels Beyond 51x51 using Sparsity [103.62784587778037]
  Recently, a couple of advanced convolutional models strike back with large kernels motivated by the local but large attention mechanism. We propose Sparse Large Kernel Network (SLaK), a pure CNN architecture equipped with 51x51 kernels that can perform on par with or better than state-of-the-art hierarchical Transformers.
  arXiv  Detail & Related papers  (2022-07-07T23:55:52Z)
• Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs [148.0476219278875]
  We revisit large kernel design in modern convolutional neural networks (CNNs) Inspired by recent advances of vision transformers (ViTs), in this paper, we demonstrate that using a few large convolutional kernels instead of a stack of small kernels could be a more powerful paradigm. We propose RepLKNet, a pure CNN architecture whose kernel size is as large as 31x31, in contrast to commonly used 3x3.
  arXiv  Detail & Related papers  (2022-03-13T17:22:44Z)
• ConTNet: Why not use convolution and transformer at the same time? [28.343371000297747]
  We propose ConTNet, combining transformer with ConvNet architectures to provide large receptive fields. We present its superiority and effectiveness on image classification and downstream tasks. We hope that ConTNet could serve as a useful backbone for CV tasks and bring new ideas for model design.
  arXiv  Detail & Related papers  (2021-04-27T22:29:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.