Research map / Detection and segmentation / CNN features
Diverse Branch Block: Building a Convolution as an Inception-Like Unit
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 regular convolution layers with DBB | VGG-16 · CIFAR-10 | top-1 accuracy: 93.95±0.03 -> 94.62±0.02 | Helped |
| Replace regular convolution layers with DBB | VGG-16 · CIFAR-100 | top-1 accuracy: 74.05±0.10 -> 75.72±0.07 | Helped |
| Replace regular convolution layers with DBB | AlexNet · ImageNet-1K | top-1 accuracy: 57.23 -> 59.19 | Helped |
| Replace regular convolution layers with DBB | MobileNet · ImageNet-1K | top-1 accuracy: 71.89 -> 72.88 | Helped |
| Replace regular convolution layers with DBB | ResNet-18 · ImageNet-1K | top-1 accuracy: 69.54 -> 70.99 | Helped |
| Replace regular convolution layers with DBB | ResNet-50 · ImageNet-1K | top-1 accuracy: 76.14 -> 76.71 | Helped |
| Use DBB-ResNet-18 as the detection backbone | CenterNet · COCO detection | AP: 29.83 -> 30.68 | Helped |
| Use DBB-ResNet-18 as the segmentation backbone | PSPNet · Cityscapes | mIoU: 70.18 -> 71.35 | Helped |
| Add the 1×1 branch (ablation) | DBB with the other branches · ImageNet-1K | top-1 accuracy: 70.36 -> 70.99 | Helped |
| Add the 1×1-K×K branch (ablation) | DBB with the other branches · ImageNet-1K | top-1 accuracy: 70.40 -> 70.99 | Helped |
| Add the 1×1-AVG branch (ablation) | DBB with the other branches · ImageNet-1K | top-1 accuracy: 70.74 -> 70.99 | Helped |
| Replace the full DBB with two duplicate K×K branches (ablation) | ResNet-18 · ImageNet-1K | top-1 accuracy: 70.99 -> 69.81 | Hurt |
Closest papers by meaning
- Efficient ConvBN Blocks for Transfer Learning and Beyond (ICLR 2024 · similarity 0.85)
- Practical Block-Wise Neural Network Architecture Generation (CVPR 2018 · similarity 0.84)
- Rethinking the Inception Architecture for Computer Vision (CVPR 2016 · similarity 0.84)
- DECO: Unleashing the Potential of ConvNets for Query-based Detection and Segmentation (ICLR 2025 · similarity 0.83)
- Learning CNN on ViT: A Hybrid Model to Explicitly Class-specific Boundaries for Domain Adaptation (CVPR 2024 · similarity 0.83)
- MogaNet: Multi-order Gated Aggregation Network (ICLR 2024 · similarity 0.83)
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.