Research map / Image generation / GANs
Recurrent Topic-Transition GAN for Visual Paragraph Generation
Xiaodan Liang, Zhiting Hu, Hao Zhang, Chuang Gan, Eric P. Xing
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 |
|---|---|---|---|
| add adversarial discriminators (fully supervised) (ablation) | RTT-GAN (Fully-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 15.07 -> 16.87 | Improved |
| add adversarial discriminators (semi-supervised) (ablation) | RTT-GAN (Semi-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 8.96 -> 13.07 | Improved |
| add sentence discriminator (ablation) | RTT-GAN (Semi-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 10.04 -> 13.07 | Improved |
| add topic-transition discriminator (ablation) | RTT-GAN (Semi-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 12.77 -> 13.07 | Improved |
| add language attention over local phrases (ablation) | RTT-GAN (Fully-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 15.13 -> 16.87 | Improved |
| add visual and language attention (ablation) | RTT-GAN (Fully-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 14.47 -> 16.87 | Improved |
| use 50 regions instead of 10 (ablation) | RTT-GAN (Fully-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 13.26 -> 16.87 | Improved |
| use 50 regions instead of 20 (ablation) | RTT-GAN (Fully-) · Image paragraph (Stanford, 2,489 test) | CIDEr: 16.15 -> 16.87 | Improved |
| augment paragraph training data with synthesized semi-supervised paragraphs | RTT-GAN · Image paragraph (Stanford, 2,489 test) | CIDEr: 16.87 -> 20.36 | Improved |
| use RTT-GAN instead of h-RNN on video | RTT-GAN · TACoS-MultiLevel video paragraph | CIDEr: 160.2 -> 165.3 | Improved |
| add discriminators in personalized generation (human vote) (ablation) | RTT-GAN (Semi-) · Personalized paragraphs (100 images) | share of judges: 12.6% -> 40.5% | Improved |
| use hierarchical regions model instead of flat image decoding | Regions-Hierarchical · Image paragraph (Stanford, 2,489 test) | CIDEr: 11.06 -> 13.52 | Improved |
Closest papers by meaning
- Matching Visual Features to Hierarchical Semantic Topics for Image Paragraph Captioning (IJCV 2022 · similarity 0.89)
- A Hierarchical Approach for Generating Descriptive Image Paragraphs (CVPR 2017 · similarity 0.88)
- Show, Reward and Tell: Automatic Generation of Narrative Paragraph From Photo Stream by Adversarial Training (AAAI 2018 · similarity 0.88)
- Let Your Photos Talk: Generating Narrative Paragraph for Photo Stream via Bidirectional Attention Recurrent Neural Networks (AAAI 2017 · similarity 0.87)
- Diverse and Coherent Paragraph Generation from Images (ECCV 2018 · similarity 0.86)
- Video Paragraph Captioning Using Hierarchical Recurrent Neural Networks (CVPR 2016 · 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.
