Research map / Detection and segmentation / CNN features
AQD: Towards Accurate Quantized Object Detection
Check 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 |
|---|---|---|---|
| use 4-bit AQD instead of the full-precision detector | RetinaNet, ResNet-18 · COCO minival | AP: 32.3 -> 34.1 | Helped |
| use 4-bit AQD instead of the full-precision detector | RetinaNet, ResNet-34 · COCO minival | AP: 36.3 -> 37.1 | Helped |
| use 4-bit AQD instead of the full-precision detector | RetinaNet, ResNet-50 · COCO minival | AP: 37.8 -> 38.1 | Helped |
| use 3-bit AQD instead of the full-precision detector | RetinaNet, ResNet-50 · COCO minival | AP: 37.8 -> 36.9 | Hurt |
| use 2-bit AQD instead of the full-precision detector | RetinaNet, ResNet-50 · COCO minival | AP: 37.8 -> 34.8 | Hurt |
| use 2-bit AQD instead of Group-Net | FCOS, ResNet-50 · COCO minival | AP: 32.7 -> 35.4 | Helped |
| replace shared BatchNorm with multi-level BatchNorm (ablation) | 2-bit FCOS, ResNet-18 · COCO minival | AP: 26.4 -> 31.8 | Helped |
| replace shared GroupNorm with multi-level BatchNorm (ablation) | 2-bit FCOS, ResNet-18 · COCO minival | AP: 29.4 -> 31.8 | Helped |
| replace shared BatchNorm with multi-level BatchNorm (ablation) | 2-bit FCOS, ResNet-50 · COCO minival | AP: 30.3 -> 35.4 | Helped |
| replace shared GroupNorm with multi-level BatchNorm (ablation) | 2-bit FCOS, ResNet-50 · COCO minival | AP: 33.4 -> 35.4 | Helped |
| quantize the feature pyramid in addition to the backbone (ablation) | FCOS, ResNet-18 (AQD*) · COCO minival | AP: 33.8 -> 33.2 | Hurt |
| quantize the detection heads in addition to backbone and feature pyramid (ablation) | FCOS, ResNet-18 (AQD*) · COCO minival | AP: 33.2 -> 32.2 | Hurt |
Closest papers by meaning
- Fully Quantized Network for Object Detection (CVPR 2019 · similarity 0.91)
- Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector (CVPR 2024 · similarity 0.88)
- Linear Symmetric Quantization of Neural Networks for Low-precision Integer Hardware (ICLR 2020 · similarity 0.88)
- AQ-DETR: Low-Bit Quantized Detection Transformer with Auxiliary Queries (AAAI 2024 · similarity 0.88)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.86)
- Towards Efficient Verification of Quantized Neural Networks (AAAI 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.