BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation

28 Jan 2022  ·  Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi ·

Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks. However, most existing pre-trained models only excel in either understanding-based tasks or generation-based tasks. Furthermore, performance improvement has been largely achieved by scaling up the dataset with noisy image-text pairs collected from the web, which is a suboptimal source of supervision. In this paper, we propose BLIP, a new VLP framework which transfers flexibly to both vision-language understanding and generation tasks. BLIP effectively utilizes the noisy web data by bootstrapping the captions, where a captioner generates synthetic captions and a filter removes the noisy ones. We achieve state-of-the-art results on a wide range of vision-language tasks, such as image-text retrieval (+2.7% in average recall@1), image captioning (+2.8% in CIDEr), and VQA (+1.6% in VQA score). BLIP also demonstrates strong generalization ability when directly transferred to video-language tasks in a zero-shot manner. Code, models, and datasets are released at https://github.com/salesforce/BLIP.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Image-text matching CommercialAdsDataset BLIP ADD(S) AUC 83.51 # 6
Image Captioning nocaps-val-in-domain BLIP_ViT-L CIDEr 114.9 # 5
SPICE 15.2 # 5
Pre-train (#images) 129M # 8
Image Captioning nocaps-val-in-domain BLIP_CapFilt-L CIDEr 111.8 # 7
SPICE 14.9 # 7
Pre-train (#images) 129M # 8
Image Captioning nocaps-val-near-domain BLIP_ViT-L CIDEr 112.1 # 5
SPICE 14.9 # 5
Pre-train (#images) 129M # 8
Image Captioning nocaps-val-near-domain BLIP_CapFilt-L CIDEr 108.6 # 7
SPICE 14.8 # 7
Pre-train (#images) 129M # 8
Image Captioning nocaps-val-out-domain BLIP_CapFilt-L CIDEr 111.5 # 6
SPICE 14.2 # 5
Pretrain (#images) 129M # 8
Image Captioning nocaps-val-out-domain BLIP_ViT-L CIDEr 115.3 # 4
SPICE 14.4 # 4
Pretrain (#images) 129M # 8
Image Captioning nocaps-val-overall BLIP_CapFilt-L CIDEr 109.6 # 7
SPICE 14.7 # 6
Pretrain (#images) 129M # 8
Image Captioning nocaps-val-overall BLIP_ViT-L CIDEr 113.2 # 5
SPICE 14.8 # 5
Pretrain (#images) 129M # 8
Open Vocabulary Attribute Detection OVAD-Box benchmark BLIP mean average precision 24.3 # 3

Methods