Research map / Architectures and efficiency / CNN design
HetConv: Heterogeneous Kernel-Based Convolutions for Deep CNNs
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace standard convolutions with HetConv, P=4 | VGG-16 · ImageNet | top-1 accuracy (%): 71.3 -> 71.2 | No clear effect |
| Replace standard convolutions with HetConv, P=4 | ResNet-34 · ImageNet | top-1 error (%): 26.80 -> 27.00 | Hurt |
| Replace standard convolutions with HetConv, P=4 | ResNet-50 · ImageNet | top-1 error (%): 23.86 -> 23.84 | Helped |
| Use HetConv, P=4 instead of groupwise plus pointwise convolution | VGG-16 · CIFAR-10 | accuracy (%): 92.76 -> 93.93 | Helped |
| Use HetConv, P=64 instead of depthwise plus pointwise convolution | VGG-16 · CIFAR-10 | accuracy (%): 91.27 -> 93.42 | Helped |
| Increase HetConv part from P=1 to P=32 | VGG-16 · CIFAR-10 | FLOPs reduced (%): -- -> 85.54 | Helped |
| Add SE to HetConv (ablation) | VGG-16 · CIFAR-10 | accuracy (%): 93.93 -> 94.29; 93.73 -> 93.41 | Mixed |
| Add SE to HetConv (ablation) | ResNet-56 · CIFAR-10 | error (%): 6.41 -> 7.16; 6.40 -> 6.75; 6.71 -> 7.79 | Hurt |
| Add SE to HetConv (ablation) | ResNet-34 · ImageNet | top-1 error (%): 27.00 -> 26.50 | Helped |
| Add SE to HetConv (ablation) | MobileNet · CIFAR-10 | accuracy (%): 92.06 -> 92.17 | Helped |
| Replace MobileNet depthwise-pointwise layers with HetConv, P=32 | MobileNet · CIFAR-10 | accuracy (%): 91.17 -> 92.06 | Helped |
Closest papers by meaning
- Approximation and Learning with Deep Convolutional Models: a Kernel Perspective (ICLR 2022 · similarity 0.87)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · similarity 0.86)
- InceptionNeXt: When Inception Meets ConvNeXt (CVPR 2024 · similarity 0.85)
- PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution (CVPR 2024 · similarity 0.85)
- Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels (ICLR 2024 · similarity 0.85)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.