Research map / Architectures and efficiency / CNN design
Container: Context Aggregation Networks
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 |
|---|---|---|---|
| Combine learnable static convolution affinity with dynamic self-attention | Container on ImageNet · ImageNet | Top-1 accuracy: 79.9 -> 82.7 | Helped |
| Use static affinity in the first three stages to reduce dynamic attention computation (ablation) | Container-Light on ImageNet · ImageNet | Top-1 accuracy: 82.7 -> 82.0 | Hurt |
| Train with 10% of the training data rather than comparing at full data | Container-Light versus DeiT-S · ImageNet, 10% training data | Top-1 accuracy: 39.3 -> 61.8 | Helped |
| Use Container-Light rather than ResNet-50 as the RetinaNet backbone | RetinaNet · COCO | mAP: 36.5 -> 43.8 | Helped |
| Use Container-Light rather than ResNet-50 as the Mask R-CNN backbone | Mask R-CNN · COCO | box AP: 38.2 -> 45.1 | Helped |
| Use Container-Light rather than ResNet-50 as the Mask R-CNN backbone | Mask R-CNN · COCO | mask AP: 34.7 -> 41.3 | Helped |
| Use Container-Light rather than ResNet-50 as the DETR backbone | DETR · COCO | mAP: 32.3 -> 38.9 | Helped |
| Use Container-Light rather than ResNet-50 as the DDETR backbone | DDETR without multi-scale · COCO | mAP: 39.3 -> 43.0 | Helped |
| Use Container-Light rather than ResNet-50 as the SMCA backbone | SMCA without multi-scale · COCO | mAP: 41.0 -> 44.2 | Helped |
| Use Container-Light rather than DeiT under DINO self-supervised training | DINO · DINO self-supervised learning | top-10 kNN accuracy: 69.6 -> 71.5 | Helped |
| Use self-attention in the hierarchical architecture (ablation) | Hierarchical DeiT-S · ImageNet | Top-1 accuracy: +1.2 | Helped |
| Replace standard convolution with group convolution (ablation) | Container architecture · ImageNet | Top-1 accuracy: 79.6 -> 79.7 | Helped |
Closest papers by meaning
- Deep Layer Aggregation (CVPR 2018 · similarity 0.89)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.89)
- Image as Set of Points (ICLR 2023 · similarity 0.87)
- Data-Driven Neuron Allocation for Scale Aggregation Networks (CVPR 2019 · similarity 0.87)
- Context Encoders: Feature Learning by Inpainting (CVPR 2016 · similarity 0.87)
- LSNet: See Large, Focus Small (CVPR 2025 · 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.