 |
ML Newsnew | past | best | rss | submit | about |
| CogVLM: Visual Expert for Pretrained Language Models (arxiv.org) | |
| 2 points by wawayanda on Nov 7, 2023 | hide | past | pdf | discuss on HN | |
|
In plain words: Instead of turning a picture into text-like inputs, this model adds trainable vision parts inside the language model's core layers, so images and words mix deeply without hurting its language skills. It topped 10 vision-and-language tests and matched a model three times its size. AbstractWe introduce CogVLM, a powerful open-source visual language foundation model. Different from the popular shallow alignment method which maps image features into the input space of language model, CogVLM bridges the gap between the frozen pretrained language model and image encoder by a trainable visual expert module in the attention and FFN layers. As a result, CogVLM enables deep fusion of vision language features without sacrificing any performance on NLP tasks. CogVLM-17B achieves state-of-the-art performance on 10 classic cross-modal benchmarks, including NoCaps, Flicker30k captioning, RefCOCO, RefCOCO+, RefCOCOg, Visual7W, GQA, ScienceQA, VizWiz VQA and TDIUC, and ranks the 2nd on VQAv2, OKVQA, TextVQA, COCO captioning, etc., surpassing or matching PaLI-X 55B. Codes and checkpoints are available at https://github.com/THUDM/CogVLM. | |
| add comment on HN |