Research map / Image generation / Text-to-image diffusion
TIME: Text and Image Mutual-Translation Adversarial Networks
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 |
|---|---|---|---|
| Add image-captioning text decoder Dec (ablation) | TIME · CUB | Inception Score: 4.58 ± 0.02 -> 4.64 ± 0.03 | Helped |
| Add image-captioning text decoder Dec (ablation) | TIME · CUB | R-precision: 69.72 ± 1.43 -> 70.72 ± 1.43 | Helped |
| Remove sentence-level embedding (ablation) | TIME · CUB | Inception Score: 4.64 ± 0.03 -> 4.64 ± 0.06 | No clear effect |
| Add 2-D positional encoding (ablation) | TIME · CUB | R-precision: 70.72 ± 1.43 -> 71.58 ± 2.39 | Helped |
| Add 2-D positional encoding (ablation) | TIME · CUB | Inception Score: 4.64 ± 0.03 -> 4.72 ± 0.06 | Helped |
| Add annealed image-text hinge loss (ablation) | TIME · CUB | Inception Score: 4.64 ± 0.03 -> 4.91 ± 0.03 | Helped |
| Add annealed image-text hinge loss (ablation) | TIME · CUB | R-precision: 70.72 ± 1.43 -> 71.57 ± 1.23 | Helped |
| Use 4 attention heads and 2 Transformer layers instead of AttnGAN attention (ablation) | AttnGAN-based generator · CUB | Inception Score: 4.36 ± 0.03 -> 4.48 ± 0.03 | Helped |
| Increase the attention configuration to 8 heads and 4 layers (ablation) | AttnGAN-based generator · CUB | Inception Score: 4.48 ± 0.03 -> 4.28 ± 0.03 | Hurt |
| Use 4 attention heads and 4 Transformer layers instead of AttnGAN attention (ablation) | AttnGAN-based generator · COCO | Inception Score: 25.89 -> 27.85 | Helped |
| Stop training the hinge loss after 400 epochs (ablation) | TIME · CUB | Inception Score | Hurt |
| Start training the hinge loss after 100 epochs (ablation) | TIME · CUB | Inception Score | Hurt |
Closest papers by meaning
- DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · similarity 0.89)
- Adversarial Learning of Semantic Relevance in Text to Image Synthesis (AAAI 2019 · similarity 0.87)
- Text-Adaptive Generative Adversarial Networks: Manipulating Images with Natural Language (NeurIPS 2018 · similarity 0.87)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · similarity 0.86)
- Leveraging Unpaired Data for Vision-Language Generative Models via Cycle Consistency (ICLR 2024 · similarity 0.86)
- MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis (AAAI 2025 · similarity 0.85)
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.