Research map / Architectures and efficiency / Quantization
INSTA-BNN: Binary Neural Network with INSTAnce-aware Threshold
Changhun Lee, Hyungjun Kim, Eunhyeok Park, Jae-Joon Kim
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 |
|---|---|---|---|
| replace RSign/RPReLU with INSTA-Th and INSTA-PReLU (ablation) | ReActNet-ResNet18 · ImageNet-1k val (from scratch ablation) | top-1 accuracy: 60.0 -> 62.1 | Improved |
| add instance-wise mean in threshold (ablation) | ReActNet-ResNet18 · ImageNet-1k val | top-1 accuracy: 60.0 -> 60.6 | Improved |
| add variance and skewness via E[x^3] (ablation) | INSTA-Th · ImageNet-1k val | top-1 accuracy: 60.6 -> 61.1 | Improved |
| add normalization layer before computing instance statistics (ablation) | INSTA-Th · ImageNet-1k val | top-1 accuracy: 60.8 -> 61.1 | Improved |
| add SE-like module for alpha (ablation) | INSTA-Th · ImageNet-1k val | top-1 accuracy: 61.1 -> 61.7 | Improved |
| use SE module alone on baseline instead of INSTA modules (ablation) | ReActNet-ResNet18 · ImageNet-1k val | top-1 accuracy: 60.0 -> 60.8 | Improved |
| skip INSTA modules in earlier layers (conv2_x) (ablation) | INSTA-BNN-ResNet18 · ImageNet-1k val | top-1 accuracy: 62.1 -> 62.1 | No clear effect |
| apply INSTA-BNN to ReActNet-A MobileNetV1 backbone | ReActNet-A · ImageNet-1k val | top-1 accuracy: 69.4 -> 71.7 | Improved |
| apply INSTA-BNN to ReActNet-ResNet18 backbone | ReActNet-ResNet18 · ImageNet-1k val | top-1 accuracy: 65.5 -> 67.6 | Improved |
| add INSTA modules on device latency | ReActNet-ResNet18 · Pixel 3 A75 latency | latency (ms): 32 -> 34 | Hurt |
| use mean, variance and skewness in threshold instead of mean only (ablation) | binary ResNet-20 · CIFAR-10 | top-1 accuracy: 87.32% is about 0.9% higher than ReActNet learnable threshold baseline | Improved |
Later papers that compared with this method
1 later accepted paper reports INSTA-BNN as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks? (AAAI 2026 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AdaBin — AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets (ECCV 2022) · better on 2 of 2 shared results
- ReCU — ReCU: Reviving the Dead Weights in Binary Neural Networks (ICCV 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
- SA-BNN: State-Aware Binary Neural Network (AAAI 2021 · similarity 0.88)
- AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets (ECCV 2022 · similarity 0.87)
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks? (AAAI 2026 · similarity 0.87)
- Sub-Bit Neural Networks: Learning To Compress and Accelerate Binary Neural Networks (ICCV 2021 · similarity 0.86)
- A&B BNN: Add&Bit-Operation-Only Hardware-Friendly Binary Neural Network (CVPR 2024 · similarity 0.86)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · 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.
