Research map / Detection and segmentation / Weakly supervised detection and segmentation
ImaginaryNet: Learning Object Detectors without Real Images and Annotations
Read the paperCheck 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 CLIP-based detection with ImaginaryNet trained on generated images and class labels | PASCAL VOC 2007 · PASCAL VOC 2007 | mAP: 20.84 -> 33.23 | Helped |
| add imaginary samples to W2N | W2N · PASCAL VOC 2007 | mAP: 56.95 -> 64.45 (+7.50) | Helped |
| add 10K imaginary data to supervised training with 5K annotated VOC2007 images | Faster R-CNN / SSOD · PASCAL VOC 2007 | mAP: 76.00 -> 80.59 (+4.49) | Helped |
| add 10K imaginary data alongside 10K unlabeled VOC2012 images | Unbiased Teacher · PASCAL VOC 2007 | mAP: 81.08 -> 82.00 (+6.00) | Helped |
| increase imaginary training samples from 2K to 10K (ablation) | ImaginaryNet (ISOD) · PASCAL VOC 2007 | mAP: 29.33 -> 35.20 | Helped |
| add a language model to generate varied scene descriptions (ablation) | ImaginaryNet (ISOD) · PASCAL VOC 2007 | mAP: 31.02 -> 33.23 | Helped |
| select imaginary classes for data ensemble (ablation) | ImaginaryNet + OICR · PASCAL VOC 2007 | mAP: 49.38 (+1.45) -> 53.05 (+5.12) | Helped |
| use DALL·E-mini instead of Stable Diffusion to generate images (ablation) | ImaginaryNet (ISOD) · PASCAL VOC 2007 | mAP: 23.28 -> 33.23 | Helped |
| enable training augmentation (ablation) | ImaginaryNet (SSOD) · PASCAL VOC 2007 | mAP: 79.92 -> 80.59 | Helped |
| enable test augmentation after training augmentation (ablation) | ImaginaryNet (SSOD) · PASCAL VOC 2007 | mAP: 80.59 -> 82.70 | Helped |
| vary language-model maximum description length (ablation) | ImaginaryNet (ISOD) · PASCAL VOC 2007 | mAP | No clear effect |
Closest papers by meaning
- Sylph: A Hypernetwork Framework for Incremental Few-Shot Object Detection (CVPR 2022 · similarity 0.86)
- Open-Vocabulary Object Detection Using Captions (CVPR 2021 · similarity 0.85)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.85)
- InstaGen: Enhancing Object Detection by Training on Synthetic Dataset (CVPR 2024 · similarity 0.85)
- Zero-shot Generalizable Incremental Learning for Vision-Language Object Detection (NeurIPS 2024 · similarity 0.85)
- Open-Vocabulary Object Detection via Language Hierarchy (NeurIPS 2024 · similarity 0.84)
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.