Papers › Florence: A New Foundation Model for Computer Vision

Florence: A New Foundation Model for Computer Vision

22 Nov 2021arXiv:2111.11432archive 2025-07-28

Lu Yuan, Dongdong Chen, Yi-Ling Chen, Noel Codella, Xiyang Dai, Jianfeng Gao, Houdong Hu, Xuedong Huang, Boxin Li, Chunyuan Li, Ce Liu, Mengchen Liu, Zicheng Liu, Yumao Lu, Yu Shi, Lijuan Wang, JianFeng Wang, Bin Xiao, Zhen Xiao, Jianwei Yang, Michael Zeng, Luowei Zhou, Pengchuan Zhang

Automated visual understanding of our diverse and open world demands computer vision models to generalize well with minimal customization for specific tasks, similar to human vision. Computer vision foundation models, which are trained on diverse, large-scale dataset and can be adapted to a wide range of downstream tasks, are critical for this mission to solve real-world computer vision applications. While existing vision foundation models such as CLIP, ALIGN, and Wu Dao 2.0 focus mainly on mapping images and textual representations to a cross-modal shared representation, we introduce a new computer vision foundation model, Florence, to expand the representations from coarse (scene) to fine (object), from static (images) to dynamic (videos), and from RGB to multiple modalities (caption, depth). By incorporating universal visual-language representations from Web-scale image-text data, our Florence model can be easily adapted for various computer vision tasks, such as classification, retrieval, object detection, VQA, image caption, video retrieval and action recognition. Moreover, Florence demonstrates outstanding performance in many types of transfer learning: fully sampled fine-tuning, linear probing, few-shot transfer and zero-shot transfer for novel images and objects. All of these properties are critical for our vision foundation model to serve general purpose vision tasks. Florence achieves new state-of-the-art results in majority of 44 representative benchmarks, e.g., ImageNet-1K zero-shot classification with top-1 accuracy of 83.74 and the top-5 accuracy of 97.18, 62.4 mAP on COCO fine tuning, 80.36 on VQA, and 87.8 on Kinetics-600.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

microsoft/unicl mentioned on GitHubpytorchMIT report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Action ClassificationAction RecognitionAction Recognition In VideosCross-Modal RetrievalImage ClassificationObject DetectionRetrievalTransfer LearningVideo RetrievalVisual Question AnsweringVisual Question Answering (VQA)Zero-Shot Cross-Modal RetrievalZero-Shot LearningZero-Shot Transfer Image ClassificationZero-Shot Transfer Image Classification (CN)Zero-Shot Video Retrievalmodelobject-detection

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Action Classification Kinetics-600 Florence (curated FLD-900M pretrain) Top-1 Accuracy 87.8 #25 of 65 Archive leaderboard report
Action Classification Kinetics-600 Florence (curated FLD-900M pretrain) Top-5 Accuracy 97.9 #25 of 65 Archive leaderboard report
Action Recognition In Videos Kinetics-400 Florence Top-1 Accuracy 86.5 #1 of 3 Archive leaderboard report
Action Recognition In Videos Kinetics-400 Florence Top-5 Accuracy 97.3 #1 of 3 Archive leaderboard report
Action Recognition In Videos Kinetics-600 Florence Top-1 Accuracy 87.8 #1 of 1 Archive leaderboard report
Action Recognition In Videos Kinetics-600 Florence Top-5 Accuracy 97.8 #1 of 1 Archive leaderboard report
Cross-Modal Retrieval COCO 2014 Florence Image-to-text R@1 81.8 #10 of 36 Archive leaderboard report
Cross-Modal Retrieval COCO 2014 Florence Image-to-text R@5 95.2 #10 of 36 Archive leaderboard report
Cross-Modal Retrieval COCO 2014 Florence Text-to-image R@1 63.2 #10 of 36 Archive leaderboard report
Cross-Modal Retrieval COCO 2014 Florence Text-to-image R@5 85.7 #10 of 36 Archive leaderboard report
Image Classification ImageNet Florence-CoSwin-H Number of params 893M #9 of 1060 Archive leaderboard report
Image Classification ImageNet Florence-CoSwin-H Top 1 Accuracy 90.05% #9 of 1060 Archive leaderboard report
Image Classification ImageNet Florence-CoSwin-H Top 5 Accuracy 99.02 #9 of 1060 Archive leaderboard report
Object Detection COCO minival Florence-CoSwin-H box AP 62 #16 of 220 Archive leaderboard report
Object Detection COCO test-dev Florence-CoSwin-H box mAP 62.4 #20 of 225 Archive leaderboard report
Video Retrieval MSR-VTT-1kA Florence text-to-video R@1 37.6 #44 of 63 Archive leaderboard report
Video Retrieval MSR-VTT-1kA Florence text-to-video R@10 72.6 #44 of 63 Archive leaderboard report
Video Retrieval MSR-VTT-1kA Florence text-to-video R@5 63.8 #44 of 63 Archive leaderboard report
Visual Question Answering VQA v2 test-dev Florence Accuracy 80.16 #7 of 11 Archive leaderboard report
Visual Question Answering VQA v2 test-std Florence overall 80.36 #3 of 3 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval COCO 2014 Florence Image-to-text R@1 64.7 #11 of 18 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval COCO 2014 Florence Image-to-text R@5 85.9 #11 of 18 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval COCO 2014 Florence Text-to-image R@1 47.2 #11 of 18 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval COCO 2014 Florence Text-to-image R@5 71.4 #11 of 18 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Image-to-text R@1 90.9 #9 of 22 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Image-to-text R@10 - #9 of 22 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Image-to-text R@5 99.1 #9 of 22 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Text-to-image R@1 76.7 #9 of 22 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Text-to-image R@10 - #9 of 22 Archive leaderboard report
Zero-Shot Cross-Modal Retrieval Flickr30k Florence Text-to-image R@5 93.6 #9 of 22 Archive leaderboard report
Zero-Shot Video Retrieval MSR-VTT Florence text-to-video R@1 37.6 #15 of 41 Archive leaderboard report
Zero-Shot Video Retrieval MSR-VTT Florence text-to-video R@10 72.6 #15 of 41 Archive leaderboard report
Zero-Shot Video Retrieval MSR-VTT Florence text-to-video R@5 63.8 #15 of 41 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: Florence

ALIGNCLIPFlorence

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections