Research map / Detection and segmentation / Open-set and few-shot detection
GTNet: Generative Transfer Network for Zero-Shot Object Detection
Shizhen Zhao, Changxin Gao, Yuanjie Shao, Lerenhan Li, Changqian Yu, Zhong Ji, Nong Sang
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 the visual-to-semantic mapping baseline with GTNet feature synthesis and unseen-class classification | GTNet · ILSVRC-2017 | mAP: 24.1 -> 26.0 | Improved |
| replace the visual-to-semantic mapping baseline with GTNet feature synthesis and unseen-class classification | GTNet · MSCOCO, IoU 0.5 | Recall@100: 34.3 -> 44.6 | Improved |
| replace the visual-to-semantic mapping baseline with GTNet feature synthesis and unseen-class classification | GTNet · Visual Genome, IoU 0.5 | Recall@100: 7.2 -> 11.3 | Improved |
| add CFU-generated unseen-class features (ablation) | GTNet · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 18.1 -> 27.0; 3.0 -> 3.6; 16.3 -> 20.1 | Improved |
| add FFU to synthesize foreground features with IoU variance (ablation) | GTNet with CFU · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 27.0 -> 39.2; 3.6 -> 7.5; 20.1 -> 24.5 | Improved |
| add BFU-generated class-specific background features (ablation) | GTNet with CFU + FFU · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 39.2 -> 43.4; 7.5 -> 10.6; 24.5 -> 25.5 | Improved |
| add classification loss L_cls (ablation) | IoUGAN · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 43.4 -> 44.3; 10.6 -> 10.9; 25.5 -> 25.7 | Improved |
| add embedding loss L_emb (ablation) | IoUGAN · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 43.4 -> 44.9; 10.6 -> 10.8; 25.5 -> 25.8 | Improved |
| add both classification and embedding losses (ablation) | IoUGAN · MSCOCO, Visual Genome, and ILSVRC-2017 | Recall@100 / mAP: 43.4 -> 44.6; 10.6 -> 11.3; 25.5 -> 26.0 | Mixed |
| use GTNet instead of the baseline detector | GTNet · MSCOCO GZSD | performance: 20.3 -> 42.5 (seen); 8.9 -> 30.4 (unseen) | Improved |
Closest papers by meaning
- Zero-Shot Object Detection (ECCV 2018 · similarity 0.89)
- Generative Dual Adversarial Network for Generalized Zero-Shot Learning (CVPR 2019 · similarity 0.87)
- Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · similarity 0.87)
- Generative Zero-Shot Learning via Low-Rank Embedded Semantic Dictionary (TPAMI 2018 · similarity 0.87)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.86)
- SAUI: Scale-Aware Unseen Imagineer for Zero-Shot Object Detection (AAAI 2024 · 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.
