Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2502.05178 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916603508031488 |
|---|---|
| author | Zhao, Yue Xue, Fuzhao Reed, Scott Fan, Linxi Zhu, Yuke Kautz, Jan Yu, Zhiding Krähenbühl, Philipp Huang, De-An |
| author_facet | Zhao, Yue Xue, Fuzhao Reed, Scott Fan, Linxi Zhu, Yuke Kautz, Jan Yu, Zhiding Krähenbühl, Philipp Huang, De-An |
| contents | We introduce Quantized Language-Image Pretraining (QLIP), a visual tokenization method that combines state-of-the-art reconstruction quality with state-of-the-art zero-shot image understanding. QLIP trains a binary-spherical-quantization-based autoencoder with reconstruction and language-image alignment objectives. We are the first to show that the two objectives do not need to be at odds. We balance the two loss terms dynamically during training and show that a two-stage training pipeline effectively mixes the large-batch requirements of image-language pre-training with the memory bottleneck imposed by the reconstruction objective. We validate the effectiveness of QLIP for multimodal understanding and text-conditioned image generation with a single model. Specifically, QLIP serves as a drop-in replacement for the visual encoder for LLaVA and the image tokenizer for LlamaGen with comparable or even better performance. Finally, we demonstrate that QLIP enables a unified mixed-modality auto-regressive model for understanding and generation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_05178 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | QLIP: Text-Aligned Visual Tokenization Unifies Auto-Regressive Multimodal Understanding and Generation Zhao, Yue Xue, Fuzhao Reed, Scott Fan, Linxi Zhu, Yuke Kautz, Jan Yu, Zhiding Krähenbühl, Philipp Huang, De-An Computer Vision and Pattern Recognition We introduce Quantized Language-Image Pretraining (QLIP), a visual tokenization method that combines state-of-the-art reconstruction quality with state-of-the-art zero-shot image understanding. QLIP trains a binary-spherical-quantization-based autoencoder with reconstruction and language-image alignment objectives. We are the first to show that the two objectives do not need to be at odds. We balance the two loss terms dynamically during training and show that a two-stage training pipeline effectively mixes the large-batch requirements of image-language pre-training with the memory bottleneck imposed by the reconstruction objective. We validate the effectiveness of QLIP for multimodal understanding and text-conditioned image generation with a single model. Specifically, QLIP serves as a drop-in replacement for the visual encoder for LLaVA and the image tokenizer for LlamaGen with comparable or even better performance. Finally, we demonstrate that QLIP enables a unified mixed-modality auto-regressive model for understanding and generation. |
| title | QLIP: Text-Aligned Visual Tokenization Unifies Auto-Regressive Multimodal Understanding and Generation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2502.05178 |