Research map / Detection and segmentation / Open-set and few-shot detection
Sylph: A Hypernetwork Framework for Incremental Few-Shot 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 |
|---|---|---|---|
| Replace the ONCE*-like hypernetwork setup with Sylph's design | iFSD · LVIS-v1 eval, Default | AP: 12.9 ($\pm$0.65) -> 18.5 ($\pm$0.12) | Helped |
| Replace the ONCE*-like hypernetwork setup with Sylph's design | iFSD · LVIS-v1 eval, Aug | AP: 8.5 ($\pm0.24) -> 20.7 ($\pm0.10) | Helped |
| Replace the ONCE*-like hypernetwork setup with Sylph's design | iFSD · LVIS-v1 eval, All | AP: 19.4 ($\pm0.08) -> 24.6 ($\pm0.10) | Helped |
| Use Sylph instead of ONCE | few-shot detection · COCO minival, 10-shot | AP_n: 1.2 -> 1.6 ($\pm$0.06) | Helped |
| Meta-train Sylph on LVIS rather than only the COCO base classes | Sylph · COCO minival, 10-shot | AP_n: 1.6 ($\pm$0.06) -> 3.8 ($\pm$0.20) | Helped |
| Add predicted bias to the baseline hypernetwork (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 12.9 ($\pm$0.65) -> 9.7 ($\pm$0.25) | Hurt |
| Add GroupNorm and weight scaling g (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 9.7 ($\pm$0.25) -> 17.2 ($\pm$0.12) | Helped |
| Use L2 weight normalization rather than GroupNorm, with bias and g (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 17.2 ($\pm$0.12) -> 18.0 ($\pm$0.12) | Helped |
| Add both GroupNorm and L2 weight normalization (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 18.0 ($\pm$0.12) -> 18.0 ($\pm$0.12) | No clear effect |
| Increase CPH depth from 1 to 2 convolutional layers (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 18.5 ($\pm$0.12) -> 18.6 ($\pm$0.13) | No clear effect |
| Remove scaling factor g from the 2-convolution configuration (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 18.6 ($\pm$0.13) -> 18.5 ($\pm$0.07) | No clear effect |
| Remove predicted bias from the 2-convolution configuration (ablation) | Sylph · LVIS, Default, modeling ablation | AP: 18.6 ($\pm$0.13) -> 18.4 ($\pm$0.07) | No clear effect |
Closest papers by meaning
- Incremental Few-Shot Object Detection (CVPR 2020 · similarity 0.91)
- Incremental-DETR: Incremental Few-Shot Object Detection via Self-Supervised Learning (AAAI 2023 · similarity 0.90)
- StarNet: towards Weakly Supervised Few-Shot Object Detection (AAAI 2021 · similarity 0.88)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.87)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.87)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · 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.