L-Verse: Bidirectional Generation Between Image and Text (2111.11133v11)
Abstract: Far beyond learning long-range interactions of natural language, transformers are becoming the de-facto standard for many vision tasks with their power and scalability. Especially with cross-modal tasks between image and text, vector quantized variational autoencoders (VQ-VAEs) are widely used to make a raw RGB image into a sequence of feature vectors. To better leverage the correlation between image and text, we propose L-Verse, a novel architecture consisting of feature-augmented variational autoencoder (AugVAE) and bidirectional auto-regressive transformer (BiART) for image-to-text and text-to-image generation. Our AugVAE shows the state-of-the-art reconstruction performance on ImageNet1K validation set, along with the robustness to unseen images in the wild. Unlike other models, BiART can distinguish between image (or text) as a conditional reference and a generation target. L-Verse can be directly used for image-to-text or text-to-image generation without any finetuning or extra object detection framework. In quantitative and qualitative experiments, L-Verse shows impressive results against previous methods in both image-to-text and text-to-image generation on MS-COCO Captions. We furthermore assess the scalability of L-Verse architecture on Conceptual Captions and present the initial result of bidirectional vision-language representation learning on general domain.
- Taehoon Kim (30 papers)
- Gwangmo Song (3 papers)
- Sihaeng Lee (8 papers)
- Sangyun Kim (4 papers)
- Yewon Seo (1 paper)
- Soonyoung Lee (10 papers)
- Seung Hwan Kim (15 papers)
- Honglak Lee (174 papers)
- Kyunghoon Bae (17 papers)