CogVLM: Visual Expert for Pretrained Language Models

We 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.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Visual Question Answering (VQA) InfiMM-Eval CogVLM-Chat Overall score 37.16 # 4
Deductive 36.75 # 4
Abductive 47.88 # 4
Analogical 28.75 # 3
Params 17B # 1
Visual Question Answering MM-Vet CogVLM(Vicuna-13B) GPT-4 score 56.8 # 5
Params 30B # 1
Visual Question Answering MM-Vet CogVLM(Vicuna-7B) GPT-4 score 52.8 # 6
Params 17B # 1

Methods


No methods listed for this paper. Add relevant methods here