Research map / Architectures and efficiency / CNN design
Towards Accurate Binary Neural Networks via Modeling Contextual Dependencies
Xingrun Xing, Yangguang Li, Wei Li, Wenrui Ding, Yalong Jiang, Yufeng Wang, Jing Shao, Chunlei Liu, Xianglong Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add binary MLP blocks replacing last conv blocks | ReActNet-A · ImageNet-1K val | top-1 accuracy: 69.4 -> 71.8 | Improved |
| add contextual dynamic embeddings to binary convolutions (ablation) | BCDNet-A · ImageNet-1K val | top-1 accuracy: 71.8 -> 72.3 | Improved |
| add binary MLP blocks to ResNet-18 ReActNet (ablation) | ReActNet (R18) · ImageNet-1K val (ResNet-18) | top-1 accuracy: 65.5 -> 66.9 | Improved |
| add dynamic embeddings on R18 (ablation) | BCDNet-A (R18) · ImageNet-1K val (ResNet-18) | top-1 accuracy: 66.9 -> 67.9 | Improved |
| use BCDNet-A instead of ReActNet-A, from scratch | ReActNet-A · Fine-grained average | accuracy: 23.2 -> 24.7 | Improved |
| add ImageNet pretraining and binary MLPs (ablation) | ReActNet-A · Fine-grained average (pretrained) | accuracy: 78.9 -> 84.2 | Improved |
| BCDNet-A vs ReActNet-A on Flowers and Aircraft from scratch | ReActNet-A · Flowers, Aircraft (from scratch) | accuracy: 29.8 -> 25.0 | Hurt |
| replace conv blocks with MLPs only in last stage (ablation) | ReActNet-18 · ImageNet-1K val | top-1 accuracy: 61.89 -> 63.58 | Improved |
| replace convs in all four stages instead of last only (ablation) | BCDNet-18 · ImageNet-1K val | top-1 accuracy: 63.58 -> 63.10 | Hurt |
| use 1 BiMLP instead of conv in last stage (ablation) | ReActNet-18 · ImageNet-1K val | top-1 accuracy: 61.89 -> 61.32 | Hurt |
| use 3 BiMLPs instead of 1 (ablation) | BCDNet-18 · ImageNet-1K val | top-1 accuracy: 61.32 -> 63.58 | Improved |
| add short-range branch (P-S-L vs P-L-L) (ablation) | BCDNet · ImageNet-1K val 600K | top-1 accuracy: 70.79 -> 71.76 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SA-BNN-50 — SA-BNN: State-Aware Binary Neural Network (AAAI 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks? (AAAI 2026 · similarity 0.87)
- Training binary neural networks with real-to-binary convolutions (ICLR 2020 · similarity 0.87)
- Network Binarization via Contrastive Learning (ECCV 2022 · similarity 0.86)
- BiMLP: Compact Binary Architectures for Vision Multi-Layer Perceptrons (NeurIPS 2022 · similarity 0.86)
- A Main/Subsidiary Network Framework for Simplifying Binary Neural Networks (CVPR 2019 · similarity 0.85)
- Gradient Matters: Designing Binarized Neural Networks via Enhanced Information-Flow (TPAMI 2021 · 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.
