Research map / Image generation / Text-to-image diffusion
L-Verse: Bidirectional Generation Between Image and Text
Taehoon Kim, Gwangmo Song, Sihaeng Lee, Sangyun Kim, Yewon Seo, Soonyoung Lee, Seung Hwan Kim, Honglak Lee, Kyunghoon Bae
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 cross-level feature augmentation to the VQ-VAE codebook | AugVAE · ImageNet1K validation | FID: 3.28 | Improved |
| Use multi-level instead of single-level feature-augmented encoding (ablation) | AugVAE · ImageNet1K validation | FID: 3.28 -> 1.04 | Improved |
| Use 32-token truncated captions instead of full 64-token captions (ablation) | L-Verse image captioning · MS-COCO Karpathy test | CIDEr: 181.6 -> 102.2 | Hurt |
| Use 32-token truncated captions instead of full 64-token captions (ablation) | L-Verse image captioning · MS-COCO Karpathy test | SPICE: 28.9 -> 23.3 | Hurt |
| Use full 64-token captions instead of 32-token truncation (ablation) | L-Verse image captioning · MS-COCO Karpathy test | CIDEr: 102.2 -> 181.6 | Improved |
| Use full 64-token captions instead of 32-token truncation (ablation) | L-Verse image captioning · MS-COCO Karpathy test | SPICE: 23.3 -> 28.9 | Improved |
| Train on Conceptual Captions rather than MS-COCO | L-Verse text-to-image · MS-COCO validation, 30,000 captions | FID-8: 29.8 -> 21.1 | Improved |
| Use L-Verse rather than prior MS-COCO-trained methods | Text-to-image generation · MS-COCO validation, 30,000 captions, FID-2 | FID-2: 55.9 -> 35.5 | Improved |
| Use L-Verse rather than prior MS-COCO-trained methods | Image captioning · MS-COCO Karpathy test | BLEU-4: 39.1 -> 39.9 | Improved |
| Use L-Verse rather than prior MS-COCO-trained methods | Image captioning · MS-COCO Karpathy test | CIDEr: 131.2 -> 102.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- DF-GAN — DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022) · better on 3 of 4 shared results
- DM-GAN — DM-GAN: Dynamic Memory Generative Adversarial Networks for Text-To-Image Synthesis (CVPR 2019) · better on 0 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
- JetFormer: An autoregressive generative model of raw images and text (ICLR 2025 · similarity 0.88)
- Leveraging Unpaired Data for Vision-Language Generative Models via Cycle Consistency (ICLR 2024 · similarity 0.88)
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.88)
- CoBIT: A Contrastive Bi-directional Image-Text Generation Model (ICLR 2024 · similarity 0.87)
- MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis (AAAI 2025 · similarity 0.87)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.87)
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.
