Unified Pix Token And Word Token Generative Language Model
For multimodal AI researchers, this work offers a potential alternative to CLIP-based vision encoders for detail-sensitive tasks, though results are preliminary and incremental.
The paper proposes a generative language model that unifies pixel tokens and word tokens to improve fine-grained visual understanding, addressing limitations in detail recognition (e.g., small text/numbers) of current CLIP/SigLIP-based multimodal models. Experiments show good performance even with small models and limited data, suggesting scalability.
Since the emergence of Vision Transformer (ViT), it has been widely used in generative language model and generative visual model. Especially in the current state-of-art open source multimodal models, ViT obtained by CLIP or SigLIP method serves as the vision encoder backbone to help them acquire visual understanding capabilities. But this method leads to limitations in visual understanding for details, such as difficulty in recognizing small text or numbers in images. To address these issues, we propose a new model to unify pix token and word token into the generative language model. The new model also features with each pix of image having its own token embedding, color folding, global conditional attention approximation and image unsupervised pretraining. We conducted image unsupervised pretraining experiments using our new model to explore its potential. The experimental results show that it has good performance even in small model and with limited training data. We believe our model also conforms to the scaling law, as long as model parameters and training data increased, its performance will continue to improve.