Research map / Image generation / GANs
Semantics-Enhanced Adversarial Nets for Text-to-Image Synthesis
Hongchen Tan, Xiuping Liu, Xin Li, Yi Zhang, Baocai Yin
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 attention competition module (AC) (ablation) | AttnGAN · CUB test | Inception score: 4.31 -> 4.61 | Improved |
| add attention competition module (AC) (ablation) | AttnGAN · COCO test | Inception score: 25.56 -> 26.83 | Improved |
| add Siamese network with sliding loss (SL) (ablation) | AttnGAN · CUB test | Inception score: 4.31 -> 4.58 | Improved |
| add Siamese network with sliding loss (SL) (ablation) | AttnGAN · COCO test | Inception score: 25.56 -> 27.13 | Improved |
| add Siamese network with contrastive loss (ablation) | AttnGAN · CUB test | Inception score: 4.31 -> 4.44 | Improved |
| replace contrastive loss with sliding loss (gamma=1/2) (ablation) | AttnGAN + SiaNet · CUB test | Inception score: 4.44 -> 4.58 | Improved |
| add both ACM and SCM (ablation) | AttnGAN · CUB test | Inception score: 4.31 -> 4.67 | Improved |
| add both ACM and SCM (ablation) | AttnGAN · COCO test | Inception score: 25.56 -> 27.86 | Improved |
| increase sliding loss gamma from 1/2 to 2 (ablation) | AttnGAN + SL · CUB test | Inception score: 4.58 -> 4.41 | Hurt |
| set attention threshold alpha=1 instead of 0.005 (ablation) | AttnGAN + AC · CUB test | Inception score: 4.61 -> 4.25 | Hurt |
| use alpha 0.005 instead of 0.001 (ablation) | AttnGAN + AC · CUB test | Inception score: 4.41 -> 4.61 | Improved |
| add SEGAN modules | AttnGAN · CUB test | FID: 22.504 -> 18.167 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- AttnGAN — AttnGAN: Fine-Grained Text to Image Generation With Attentional Generative Adversarial Networks (CVPR 2018) · better on 6 of 6 shared results
- MirrorGAN — MirrorGAN: Learning Text-To-Image Generation by Redescription (CVPR 2019) · better on 2 of 2 shared results
- StackGAN — StackGAN: Text to Photo-Realistic Image Synthesis With Stacked Generative Adversarial Networks (ICCV 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Photographic Text-to-Image Synthesis With a Hierarchically-Nested Adversarial Network (CVPR 2018 · similarity 0.92)
- CPGAN: Content-Parsing Generative Adversarial Networks for Text-to-Image Synthesis (ECCV 2020 · similarity 0.92)
- Adversarial Learning of Semantic Relevance in Text to Image Synthesis (AAAI 2019 · similarity 0.90)
- Semantic Image Synthesis via Adversarial Learning (ICCV 2017 · similarity 0.90)
- DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · similarity 0.90)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · similarity 0.88)
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.
