text stringlengths 102 5.9k | source stringclasses 1
value |
|---|---|
Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks Bin Xiao†Haiping Wu∗Weijian Xu∗Xiyang Dai Houdong Hu Yumao Lu Michael Zeng Ce Liu‡Lu Yuan‡ †project lead∗equal contribution‡direcional lead Azure AI, Microsoft Abstract We introduce Florence-2, a novel vision foundation model with a unified, p... | 2311.06242v1.pdf |
This pursuit is characterized by distinctiveness and sub-stantial challenges. A key hurdle is the scarcity of com-prehensive visual annotations, hindering the development of a foundational model capable of capturing the intricate nuances of spatial hierarchy and semantic granularity. Ex-isting datasets, such as Image N... | 2311.06242v1.pdf |
Locate the objects in the image. Image Encoder Transformer Encoders Transformer Decoders What does the image describe? Locate the phrases in the caption:A woman riding a bike. What does theregion (0. 41, 0. 15, 0. 63, 0. 73)describe? What is the polygon mask of region (0. 41, 0. 15, 0. 63, 0. 73)?The image shows a pers... | 2311.06242v1.pdf |
3. Model We present the foundation model Florence-2, designed for universal representation learning, capable of handling various vision tasks with a single set of weights and a uni-fied architecture. As depicted in Figure 2, Florence-2 em-ploys a sequence-to-sequence learning paradigm [77], inte-grating all tasks, desc... | 2311.06242v1.pdf |
Segmentation Grounding Azure OCR API Segmentation Model Object / Proposal Det. Caption Model Iterative data refinement Initial annotation w/ specialists Data filtering and enhancement Region filtering Florence-2Multi-round re-training Fine-tunew/ small-scale enhanceddata Train w/ large-scale data Grounding Model Image ... | 2311.06242v1.pdf |
finement process incrementally improves the quality of our training dataset. In the case of tasks we initially bypassed due to insuf-ficient data for the training of a robust specialist model, we leveraged the iteratively trained model for pre-training purposes. Subsequent fine-tuning of this pre-trained model with the... | 2311.06242v1.pdf |
Text annotations Less granular (image level) More granular (image level) Region-text pairs annotations None semantic Rich semantic Text-phrase-region annotations Less granular (region level) More granular (region level) The image shows a person riding a red bicycle on a road with a red car in the background. The person... | 2311.06242v1.pdf |
Annotation Type Text Type #Image Annotations #Avg Tokens #Regions #Avg Regions #Avg Regional Tokens Text Brief 235M 7. 95---Detailed 126M 31. 65---More detailed 126M 70. 53---Region-Text Phrase 126M-681M 5. 42 1. 19 Brief 126M-681M 5. 42 2. 55 Text-Phrase-Region Brief 235M 7. 95 1007M 4. 27 1. 93 Detailed 126M 31. 65 1... | 2311.06242v1.pdf |
0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 square-root normalized area012345probability density Distribution of area of bounding boxes region-text text-phrase-region(a) 3 2 1 0 1 2 3 log aspect ratio0. 000. 250. 500. 751. 001. 251. 50probability density Distribution of aspect ratio of bounding boxes region-text text-phrase-region (... | 2311.06242v1.pdf |
Method #params COCO Cap. No Caps Text Caps COCO Det. Flickr30k Refcoco Refcoco+ Refcocog Refcoco RES test val val val2017 test val test-A test-B val test-A test-B val test val CIDEr CIDEr CIDEr m AP R@1 Accuracy Accuracy Accuracy m Io U Flamingo [2] 80B 84. 3-------------Kosmos-2 [60] 1. 6B----78. 7 52. 3 57. 4 47. 3 4... | 2311.06242v1.pdf |
Method #params COCO Det. Flickr30k Refcoco Refcoco+ Refcocog Refcoco RES val2017 test val test-A test-B val test-A test-B val test val m AP R@1 Accuracy Accuracy Accuracy m Io U Specialist Models Seq TR [99]---83. 7 86. 5 81. 2 71. 5 76. 3 64. 9 74. 9 74. 2-Poly Former [49]---90. 4 92. 9 87. 2 85. 0 89. 8 78. 0 85. 8 8... | 2311.06242v1.pdf |
Mask R-CNN DINO Backbone Pretrain APb APm AP Vi T-B [46] MAE, IN-1k 51. 6 45. 9 55. 0 Swin-B [51] Sup IN-1k 50. 2-53. 4 Swin-B [51] Sim MIM [83] 52. 3--Focal Att-B [86] Sup IN-1k 49. 0 43. 7-Focal Net-B [85] Sup IN-1k 49. 8 44. 1 54. 4 Conv Ne Xt v1-B [52] Sup IN-1k 50. 3 44. 9 52. 6 Conv Ne Xt v2-B [81] Sup IN-1k 51. ... | 2311.06242v1.pdf |
2500 5000 7500 10000 12500 15000 17500 20000 Optimization Steps1. 291. 301. 311. 321. 331. 341. 35CIDEr COCO Caption 2500 5000 7500 10000 12500 15000 17500 20000 Optimization Steps0. 050. 100. 150. 200. 250. 30AP COCO Object Detection 2500 5000 7500 10000 12500 15000 17500 20000 Optimization Steps0. 20. 30. 40. 50. 60.... | 2311.06242v1.pdf |
Caption Detection Grounding RES V Pre L Pre CIDEr AP Recall@1 m IOU o IOU Freeze Vision Encoder ✓ ✓ 120. 0 6. 9 66. 3 9. 9 13. 6 Unfreeze Vision Encoder ✓ 81. 3 4. 9 69. 0 15. 3 15. 6 ✓ 117. 4 19. 6 75. 2 21. 5 19. 3 ✓ ✓ 118. 7 19. 7 76. 3 18. 6 17. 8 Table 12. Basic components. Zero-shot performance on COCO caption, C... | 2311.06242v1.pdf |
decoder models [15, 55] adapted for sequence-to-sequence learning, but uses our built large-scale comprehensive anno-tation data instead of combining existing sparse annotated data. 7. 2. Vision Datasets Comprehensive annotations. The quest for comprehen-sive understanding of visual scenes, the holy grail of com-puter ... | 2311.06242v1.pdf |
Clark, Christopher Berner, Sam Mc Candlish, Alec Rad-ford, Ilya Sutskever, and Dario Amodei. Language mod-els are few-shot learners. In H. Larochelle, M. Ranzato, R. Hadsell, M. F. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems, volume 33, pages 1877-1901. Curran Associates, Inc., 2020. ... | 2311.06242v1.pdf |
[29] Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision, 2021. 9, 14, 15 [30] Andrej Karpathy and Li Fei-Fei. Deep visual-semantic alignments for generating i... | 2311.06242v1.pdf |
[53] Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts, 2017. 9 [54] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization, 2019. 9, 11 [55] Jiasen Lu, Christopher Clark, Rowan Zellers, Roozbeh Mottaghi, and Aniruddha Kembhavi. Unified-io: A unified model for visio... | 2311.06242v1.pdf |
[81] Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, and Saining Xie. Con-vnext v2: Co-designing and scaling convnets with masked autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16133-16142, 2023. 10, 11, 12 [82] Tete Xiao, Yin... | 2311.06242v1.pdf |
A. Supported Tasks and Annotations in Florence-2 Task Annotation Type Prompt Input Output Caption Text Image, text Text Detailed caption Text Image, text Text More detailed caption Text Image, text Text Region proposal Region Image, text Region Object detection Region-Text Image, text Text, region Dense region caption ... | 2311.06242v1.pdf |
D. More Examples of Annotations in FLD-5B (a) Region only (b) Region w/ phrases (c) Region w/ brief text acoupleof1beds2sitting next to each other in a room3. 1 apolicecar1parked next toother cars2on adirtroad3. 1 alberteinstein1quote2if you can ' t explain it simply, you don ' t understand it well enough. 1 awooden de... | 2311.06242v1.pdf |
(a) Region only (b) Region w/ phrases (c) Region w/ brief text know your cows1byjack2byard3. 1 agroup ofpeople1standing next to each other holding a ball2. 1 apanda bear1walking on the ground in front of a rock2. 1 a close up of aplant1with aflower2in the background. 1 (d) Text-phrase-region w/ brief text the image sho... | 2311.06242v1.pdf |
E. Qualitative Evaluation and Visualization Results E. 1. Detailed Image Caption Detailed Image Caption Prompt : Describe with a paragraph what is shown in the image. The image is a hand-drawn illustration of a glass of gin and tonic. The glass is filled with ice cubes, lime slices, and a slice of lemon. There is a str... | 2311.06242v1.pdf |
E. 2. Visual Grounding Visual Grounding Prompt : Locate the phrases in the caption: {caption } The image shows a group of five cartoon monsters. On the left side, there is abrown monster1with horns and a big smile on its face. Next to it, there are two smaller monsters2, one black and one green. The black monster3has t... | 2311.06242v1.pdf |
Visual Grounding Prompt : Locate the phrases in the caption: {caption } The image is a flat lay of various food items arranged on a white marble countertop. On the left side of the image, there is apiece ofsalmon1. Next to it, there are slices ofcheese2,aglass ofoil3, coffee beans4,azucchini5, a bunch of straw berries6... | 2311.06242v1.pdf |
E. 3. Dense Region Caption Dense Region Caption Figure 13. Dense region caption prediction results. 26 | 2311.06242v1.pdf |
E. 4. Open Vocabulary Detection Open V ocabulary Object Detection Prompt : Locate Five Alive juice box⟨and⟩Col-gate tooth paste in the image. Prompt : Locate Chew bacca in the image. Prompt : Locate giraffe in the image. Prompt : Locate Mercedes-Benz⟨and⟩M2⟨and⟩Audi in the image. Prompt : Locate the objects with catego... | 2311.06242v1.pdf |
E. 5. OCR Ocr with region Prompt : What is the text in the image, with regions? Easy Stroganoff1 Brown 1lb. ground beef inskil let2 Add 1can beef broth3 1can cream ofmush room soup4 Cut insquares &2dld toabove5 1/Boz pkg. cream cheese6 Simmer 20-3min. 7 Serve over hotrice /noo dles. 8 Vintage. Recipes/Easy-Strogano9Cha... | 2311.06242v1.pdf |
E. 6. Region to segmentation Region to Segmentation Prompt : What is the polygon mask of region ⟨loc586⟩⟨loc294⟩⟨loc929⟩⟨loc814⟩ Prompt : What is the polygon mask of region ⟨loc317⟩⟨loc314⟩⟨loc893⟩⟨loc904⟩ Prompt : What is the polygon mask of region ⟨loc541⟩⟨loc266⟩⟨loc692⟩⟨loc627⟩ Prompt : What is the polygon mask of ... | 2311.06242v1.pdf |
F. Comparision with LMMs on Detailed Caption Task Comparison with LMMs on Detailed Image Caption Task Florence-2 : The image is a photograph of a handwritten note from a person named Angelica. The note is written in black ink on a white background. The handwriting is neat and legible, and it appears to be a handwritten... | 2311.06242v1.pdf |
Comparison with LMMs on Detailed Image Caption Task Florence-2 : The image is a close-up of a red apple with a green label on it. The label has the text "FUJI 94131 USA" and a barcode on it, indicating that the apple is from the United States. The apple appears to be fresh and has a shiny texture. The background is whi... | 2311.06242v1.pdf |
Comparison with LMMs on Detailed Image Caption Task Florence-2 : The image shows a highway with multiple lanes. There are several vehicles on the road, including a black van and a red truck. On the left side of the image, there is a white semi-truck with the word "Martini" written on it. Above the highway, there are se... | 2311.06242v1.pdf |
Comparison with LMMs on Detailed Image Caption Task Florence-2 : The image shows a kitchen countertop with a laptop on it. The laptop is open and the screen is displaying a blue abstract design. On the countertop, there are various kitchen utensils hanging on a metal rack. The rack is attached to the wall above the sin... | 2311.06242v1.pdf |
G. Comarison with Kosmos-2 on Detailed Caption and Grounding Tasks Comparison on detailed caption and grounding tasks. The image features a home office with two white desks, one on the left and the other on the right. The desks are connected by awhite metal frame, creating a functional and stylish workspace. Alap top i... | 2311.06242v1.pdf |
Comparison on detailed caption and grounding tasks. The image features a white dining table with a variety of food items on it. On the table, there is a bowl of bread, a bowl with a salad, and a plate with a side of yogurt. In addition to the food, there are two cups placed on the table. One cup is located near the lef... | 2311.06242v1.pdf |
README.md exists but content is empty.
- Downloads last month
- 12