Research map / Detection and segmentation / CNN features
HBONet: Harmonious Bottleneck on Two Orthogonal Dimensions
Duo Li, Aojun Zhou, Anbang Yao
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 MobileNetV2 inverted bottlenecks with HBO modules (width 1.0) | MobileNetV2 (1.0) · ImageNet val | top-1 accuracy (%): 72.2 -> 73.1 | Improved |
| replace MobileNetV2 bottlenecks with HBO modules at width 0.25 | MobileNetV2 (0.25) · ImageNet val | top-1 accuracy (%): 52.3 -> 57.3 | Improved |
| replace MobileNetV2 bottlenecks with HBO modules at width 0.1 | MobileNetV2 (0.1) · ImageNet val | top-1 accuracy (%): 34.9 -> 41.5 | Improved |
| use HBONet at 224x224 input, width 0.8 vs 0.75 | MobileNetV2 (0.75) · ImageNet val (224 input) | top-1 accuracy (%): 69.8 -> 71.3 | Improved |
| use HBONet at 96x96 input (width 0.8) | MobileNetV2 (0.75) · ImageNet val (96 input) | top-1 accuracy (%): 58.8 -> 61.4 | Improved |
| use HBONet at 96x96 input (width 0.35) | MobileNetV2 (0.35) · ImageNet val (96 input) | top-1 accuracy (%): 45.5 -> 50.3 | Improved |
| HBONet 0.6 at 192 input instead of MobileNetV2 0.5 at 224 (similar FLOPs) | MobileNetV2 (0.5) · ImageNet val | top-1 accuracy (%): 64.6 -> 67.7 | Improved |
| HBONet 0.5 at 224 input instead of MobileNetV2 0.6 at 192 (similar FLOPs) | MobileNetV2 (0.6) · ImageNet val | top-1 accuracy (%): 65.6 -> 67.3 | Improved |
| use HBONet backbone in SSD320 at width 1.0 | MobileNetV2 SSD320 · PASCAL VOC 2007 test | mAP (%): 70.4 -> 71.0 | Improved |
| use HBONet backbone in SSD320 at width 0.1 | MobileNetV2 SSD320 · PASCAL VOC 2007 test | mAP (%): 36.3 -> 42.6 | Improved |
| use HBONet backbone for re-ID at width 1.0 | MobileNetV2 · Market-1501 | mAP (%): 70.5 -> 74.4 | Improved |
| use HBONet backbone for re-ID at width 0.1 | MobileNetV2 · Market-1501 | Rank-1 (%): 68.3 -> 73.2 | Improved |
Closest papers by meaning
- MobileNetV2: Inverted Residuals and Linear Bottlenecks (CVPR 2018 · similarity 0.88)
- MobileDets: Searching for Object Detection Architectures for Mobile Accelerators (CVPR 2021 · similarity 0.88)
- MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models (ICLR 2023 · similarity 0.88)
- MobileODE: An Extra Lightweight Network (NeurIPS 2025 · similarity 0.87)
- FreeNet: Liberating Depth-Wise Separable Operations for Building Faster Mobile Vision Architectures (AAAI 2025 · similarity 0.86)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · 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.
