Research map / Architectures and efficiency / CNN design
Deep Layer Aggregation
Read the paperCheck 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 |
|---|---|---|---|
| Use DLA-X-60-C instead of SqueezeNet-A at a similar parameter scale | Compact ImageNet classifier · ImageNet 2012 val | Top-1 error: 42.5 -> 32.0 | Helped |
| Use DLA-X-46-C instead of SqueezeNet-A at a similar parameter scale | Compact ImageNet classifier · ImageNet 2012 val | Top-1 error: 42.5 -> 34.0 | Helped |
| Reduce segmentation output stride from 8s to 2s (ablation) | DLA-34 · CamVid | mIoU: 66.7 -> 68.6 | Helped |
| Use a deeper DLA-102 at 2s rather than DLA-34 at 2s (ablation) | DLA segmentation · CamVid | mIoU: 68.6 -> 71.0 | Helped |
| Reduce segmentation output stride from 8s to 2s (ablation) | DLA-34 · Cityscapes val | mIoU: 73.5 -> 75.1 | Helped |
| Reduce boundary detection output stride from 8s to 2s (ablation) | DLA-34 · BSDS | ODS: 0.760 -> 0.794 | Helped |
| Reduce boundary detection output stride from 8s to 2s (ablation) | DLA-34 · BSDS | OIS: 0.772 -> 0.808 | Helped |
| Increase backbone depth from DLA-34 to DLA-102 at 2s (ablation) | DLA boundary detector · BSDS | ODS: 0.794 -> 0.803 | Helped |
| Increase backbone depth from DLA-34 to DLA-102 at 2s (ablation) | DLA boundary detector · BSDS | AP: 0.624 -> 0.623 | Hurt |
| Increase backbone depth from DLA-34 to DLA-102 at 2s (ablation) | DLA boundary detector · PASCAL Boundaries | ODS: 0.743 -> 0.754 | Helped |
| Add residual connections to aggregation nodes (ablation) | HDA networks with hierarchy depth of 4 levels or more · Image classification | accuracy | Helped |
| Add residual connections to aggregation nodes (ablation) | HDA networks with smaller hierarchies · Image classification | accuracy | Hurt |
Closest papers by meaning
- Container: Context Aggregation Networks (NeurIPS 2021 · similarity 0.89)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.87)
- SparX: A Sparse Cross-Layer Connection Mechanism for Hierarchical Vision Mamba and Transformer Networks (AAAI 2025 · similarity 0.87)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.86)
- LookWhere? Efficient Visual Recognition by Learning Where to Look and What to See from Self-Supervision (NeurIPS 2025 · similarity 0.86)
- ZipIt! Merging Models from Different Tasks without Training (ICLR 2024 · similarity 0.86)
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.