REVIEW 6 cited by
More ConvNets in the 2020s: Scaling up Kernels Beyond 51x51 using Sparsity
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Transformers have quickly shined in the computer vision world since the emergence of Vision Transformers (ViTs). The dominant role of convolutional neural networks (CNNs) seems to be challenged by increasingly effective transformer-based models. Very recently, a couple of advanced convolutional models strike back with large kernels motivated by the local-window attention mechanism, showing appealing performance and efficiency. While one of them, i.e. RepLKNet, impressively manages to scale the kernel size to 31x31 with improved performance, the performance starts to saturate as the kernel size continues growing, compared to the scaling trend of advanced ViTs such as Swin Transformer. In this paper, we explore the possibility of training extreme convolutions larger than 31x31 and test whether the performance gap can be eliminated by strategically enlarging convolutions. This study ends up with a recipe for applying extremely large kernels from the perspective of sparsity, which can smoothly scale up kernels to 61x61 with better performance. Built on this recipe, we propose Sparse Large Kernel Network (SLaK), a pure CNN architecture equipped with sparse factorized 51x51 kernels that can perform on par with or better than state-of-the-art hierarchical Transformers and modern ConvNet architectures like ConvNeXt and RepLKNet, on ImageNet classification as well as a wide range of downstream tasks including semantic segmentation on ADE20K, object detection on PASCAL VOC 2007, and object detection/segmentation on MS COCO.
Forward citations
Cited by 6 Pith papers
-
SWIFT: Spatio-temporal Wavelet Integrated Forecasting Framework for Workload Traces
SWIFT forecasts cloud workloads at SOTA accuracy with linear complexity by combining learnable cascaded wavelet convolutions and sequential inter/intra-variable mixing.
-
On the Reliability of Cue Conflict and Beyond
Stylized cue-conflict bias scores are confounded by impure cues, imbalance, ratio metrics and restricted labels; REFINED-BIAS supplies pure balanced stimuli and full-label MRR sensitivity for reliable diagnosis.
-
Partial Ring Scan: Revisiting Scan Order in Vision State Space Models
Ring-based scanning with selective channel routing improves accuracy, speed, and rotation robustness of vision state-space models.
-
MambaFusion: Height-Fidelity Dense Global Fusion for Multi-modal 3D Object Detection
A camera-LiDAR 3D detector built around a hybrid local-global Mamba block with height-fidelity LiDAR encoding reports 75.0 NDS on nuScenes validation, outperforming prior transformer-based fusion methods.
-
BEVANet: Bilateral Efficient Visual Attention Network for Real-Time Semantic Segmentation
BEVANet reports 81.0% mIoU at 33 FPS on Cityscapes, 0.9% above PIDNet-M but with more parameters and FLOPs.
-
Moir\'e Zero: An Efficient and High-Performance Neural Architecture for Moir\'e Removal
MZNet reports state-of-the-art moiré removal on high-resolution benchmarks with a U-Net combining multi-scale dual attention, multi-shape large-kernel convolutions, and fused skip connections.
Discussion (0). Continue with ORCID to comment.