Research map / Architectures and efficiency / CNN design
XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks
Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, Ali Farhadi
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 |
|---|---|---|---|
| compute scaling factor alpha in closed form instead of as a learned layer parameter (ablation) | BWN AlexNet · ImageNet val | top-1 accuracy: 46.2 -> 56.8 | Improved |
| compute scaling factor alpha in closed form instead of as a learned layer parameter (ablation) | BWN AlexNet · ImageNet val | top-5 accuracy: 69.5 -> 79.4 | Improved |
| reorder block to B-A-C-P instead of C-B-A-P (ablation) | XNOR-Net AlexNet · ImageNet val | top-1 accuracy: 30.3 -> 44.2 | Improved |
| reorder block to B-A-C-P instead of C-B-A-P (ablation) | XNOR-Net AlexNet · ImageNet val | top-5 accuracy: 57.5 -> 69.2 | Improved |
| add input scaling factor beta (ablation) | XNOR-Net AlexNet · ImageNet val | top-1 accuracy: less than 1% | No clear effect |
| binarize gradients in backward pass (ablation) | XNOR-Net AlexNet · ImageNet val | top-1 accuracy: drop only by 1.4% | Hurt |
| binarize weights (BWN) versus full precision | AlexNet · ImageNet val | top-1 accuracy: 56.6 -> 56.8 | No clear effect |
| BWN versus BinaryConnect | AlexNet · ImageNet val | top-1 accuracy: 35.4 -> 56.8 | Improved |
| XNOR-Net versus BinaryNet | AlexNet · ImageNet val | top-1 accuracy: 27.9 -> 44.2 | Improved |
| binarize weights (BWN) versus full precision | ResNet-18 · ImageNet val | top-1 accuracy: 69.3 -> 60.8 | Hurt |
| binarize weights and inputs (XNOR-Net) versus full precision | ResNet-18 · ImageNet val | top-1 accuracy: 69.3 -> 51.2 | Hurt |
| binarize weights (BWN) versus full precision | GoogLenet variant · ImageNet val | top-1 accuracy: 71.3 -> 65.5 | Hurt |
Later papers that compared with this method
20 later accepted papers report XNOR-Net as a baseline in their results tables; 20 of them report a better number on most of the results they share with it.
- PikeLPN: Mitigating Overlooked Inefficiencies of Low-Precision Neural Networks (CVPR 2024 · better on 2 of 2 shared results)
- Lipschitz Continuity Retained Binary Neural Network (ECCV 2022 · better on 1 of 1 shared result)
- SiMaN: Sign-to-Magnitude Network Binarization (TPAMI 2022 · better on 1 of 1 shared result)
- BATS: Binary ArchitecTure Search (ECCV 2020 · better on 2 of 2 shared results)
- ReActNet: Towards Precise Binary Neural Network with Generalized Activation Functions (ECCV 2020 · better on 1 of 1 shared result)
- Bayesian Optimized 1-Bit CNNs (ICCV 2019 · better on 5 of 5 shared results)
- A Main/Subsidiary Network Framework for Simplifying Binary Neural Networks (CVPR 2019 · better on 4 of 4 shared results)
- Regularizing Activation Distribution for Training Binarized Deep Networks (CVPR 2019 · better on 4 of 4 shared results)
12 more not listed.
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
- TBN: Convolutional Neural Network with Ternary Inputs and Binary Weights (ECCV 2018 · similarity 0.89)
- Training binary neural networks with real-to-binary convolutions (ICLR 2020 · similarity 0.89)
- Towards Accurate Binary Convolutional Neural Network (NeurIPS 2017 · similarity 0.88)
- Bi-Real Net: Binarizing Deep Network Towards Real-Network Performance (IJCV 2019 · similarity 0.88)
- Network Sketching: Exploiting Binary Structure in Deep CNNs (CVPR 2017 · similarity 0.88)
- A Main/Subsidiary Network Framework for Simplifying Binary Neural Networks (CVPR 2019 · similarity 0.87)
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.
