Xiaohan Ding, Xiangyu Zhang, Yizhuang Zhou, Jungong Han, Guiguang Ding, Jian Sun
Paper appears method- or tooling-adjacent to AI workflows with partial ecosystem coverage.
We revisit large kernel design in modern convolutional neural networks (CNNs). Inspired by recent advances in 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 suggested five guidelines, e.g., applying re-parameterized large depth-wise convolutions, to design efficient high-performance large- ...
kernel CNNs. Following the guidelines, we propose RepLKNet, a pure CNN architecture whose kernel size is as large as 31x31, in contrast to commonly used 3x3. RepLKNet greatly closes the performance gap between CNNs and ViTs, e.g., achieving comparable or superior results than Swin Transformer on ImageNet and a few typical downstream tasks, with lower latency. RepLKNet also shows nice scalability to big data and large models, obtaining 87.8% top-1 accuracy on ImageNet and 56.0% mIoU on ADE20K, which is very competitive among the state-of-the-arts with similar model sizes. Our study further reveals that, in contrast to small-kernel CNNs, large-kernel CNNs have much larger effective receptive fields and higher shape bias rather than texture bias. Code & models at https://github.com/megvii-research/RepLKNet.
Some benchmark signal exists in the extracted evidence, but it is not structured strongly enough yet for a confident benchmark decision.
We revisit large kernel design in modern convolutional neural networks (CNNs).
cmhungsteve/Awesome-Transformer-Attention is the closest maintained adjacent implementation (Matches contextual method/domain keyword: transformer). It is not paper-verified; validate algorithm and evaluation setup against the paper before trusting reported metrics. Community adoption signal: 5041 GitHub stars.
Hardware Notes
Expect multi-day setup/compute for meaningful reproduction based on current guidance.
Evidence graph: 3 refs, 3 links.
Utility signals: depth 85/100, grounding 85/100, status high.
Compare maintenance quality, reproducibility coverage, and evidence confidence before choosing a reproduction baseline.
Strong overlap with paper title keywords · Community adoption signal (944 stars)
Risk flags
There is no verified maintained implementation yet. Use this baseline plan to decide whether to prototype now or defer.
Hardware requirements
No verified implementation available
Framework baselines
Modern transformer training baseline.
Reference transformer building block implementation.
These are not paper-verified. Use them as reference points when no direct implementation is available.
Matches contextual method/domain keyword: transformer
No additional verified repositories beyond the primary recommendation.
These repositories had low-confidence matching signals and are hidden by default.
No trustworthy direct or curated related Hugging Face artifacts were found yet.
Continue with targeted Hugging Face searches derived from the paper title and method context:
Models
Tip: start with models, then check datasets/spaces if you need evaluation data or demos.
Direct artifact matches are currently sparse. Use targeted Hugging Face searches to quickly locate candidate models, datasets, and demos.
78
Citations
0
References
Tasks
Computer science, Kernel (algebra), Convolutional neural network, Parameterized complexity, Scaling, Scalability, Tree kernel, Contrast (vision)
Methods
Transformer, Algorithm
Domains
Artificial intelligence, Machine learning, Computer Vision and Pattern Recognition
Evaluation & Human Feedback Data
Open this paper in HFEPX to review benchmark signals, evaluation modes, and human-feedback protocol context.
Open in HFEPXExplore Similar Papers
Jump to Paper2Code search queries derived from this paper's research context.
Need human evaluators for your AI research? Scale annotation with expert AI Trainers.