1、VRD数据集
- VRD数据集用于图像中视觉关系的识别，数据集地址：https://cs.stanford.edu/people/ranjaykrishna/vrd/，整个数据集包括5000张图像，37993000个目标关系，70中目标关系（谓语）主要包括以下5类：
结果展示：
- 数据集包括100个目标类别以及目标对应的70个关系类别，其中，
100个目标类别为：
1th object category: person，2th object category: sky，3th object category: building，
4th object category: truck，5th object category: bus，6th object category: table，
7th object category: shirt，8th object category: chair， 9th object category: car
10th object category: train，11th object category: glasses，12th object category: tree
13th object category: boat，14th object category: hat， 15th object category: trees
16th object category: grass，17th object category: pants，18th object category: road
19th object category: motorcycle，20th object category: jacket，21th object category: monitor
22th object category: wheel，23th object category: umbrella，24th object category: plate
25th object category: bike，26th object category: clock，27th object category: bag
28th object category: shoe，29th object category: laptop，30th object category: desk
31th object category: cabinet，32th object category: counter，33th object category: bench
34th object category: shoes，35th object category: tower，36th object category: bottle
37th object category: helmet，38th object category: stove，39th object category: lamp
40th object category: coat，41th object category: bed，42th object category: dog
43th object category: mountain，44th object category: horse，45th object category: plane
46th object category: roof，47th object category: skateboard，48th object category: traffic light
49th object category: bush，50th object category: phone，51th object category: airplane
52th object category: sofa，53th object category: cup，54th object category: sink
55th object category: shelf，56th object category: box，57th object category: van
58th object category: hand，59th object category: shorts，60th object category: post
61th object category: jeans，62th object category: cat， 63th object category: sunglasses
64th object category: bowl， 65th object category: computer，66th object category: pillow
67th object category: pizza，68th object category: basket，69th object category: elephant
70th object category: kite，71th object category: sand，72th object category: keyboard
73th object category: plant，74th object category: can，75th object category: vase，
76th object category: refrigerator，77th object category: cart，78th object category: skis
79th object category: po，80th object category: surfboard，81th object category: paper
82th object category: mouse，83th object category: trash can， 84th object category: cone
85th object category: camera，86th object category: ball，87th object category: bear
88th object category: giraffe，89th object category: tie，90th object category: luggage
91th object category: faucet，92th object category: hydrant，93th object category: snowboard
94th object category: oven，95th object category: engine，96th object category: watch
97th object category: face， 98th object category: street，99th object category: ramp
100th object category: suitcase
70个关系类别为：
1th predicate category: on，2th predicate category: wear，3th predicate category: has
4th predicate category: next to，5th predicate category: sleep next to，6th predicate category: sit next to
7th predicate category: stand next to，8th predicate category: park next，9th predicate category: walk next to
10th predicate category: above，11th predicate category: behind，12th predicate category: stand behind
13th predicate category: sit behind，14th predicate category: park behind，15th predicate category: in the front of
16th predicate category: under，17th predicate category: stand under，18th predicate category: sit under
19th predicate category: near， 20th predicate category: walk to，21th predicate category: walk
22th predicate category: walk past，23th predicate category: in，24th predicate category: below
25th predicate category: beside，26th predicate category: walk beside，27th predicate category: over
28th predicate category: hold，29th predicate category: by，30th predicate category: beneath
31th predicate category: with，32th predicate category: on the top of，33th predicate category: on the left of
34th predicate category: on the right of，35th predicate category: sit on，36th predicate category: ride
37th predicate category: carry，38th predicate category: look，39th predicate category: stand on
40th predicate category: use，41th predicate category: at，42th predicate category: attach to
43th predicate category: cover，44th predicate category: touch，45th predicate category: watch
46th predicate category: against，47th predicate category: inside，48th predicate category: adjacent to
49th predicate category: across，50th predicate category: contain，51th predicate category: drive
52th predicate category: drive on，53th predicate category: taller than，54th predicate category: eat
55th predicate category: park on，56th predicate category: lying on，57th predicate category: pull
58th predicate category: talk，59th predicate category: lean on，60th predicate category: fly
61th predicate category: face，62th predicate category: play with，63th predicate category: sleep on
64th predicate category: outside of，65th predicate category: rest on，66th predicate category: follow
67th predicate category: hit，68th predicate category: feed，69th predicate category: kick
70th predicate category: skate on
2、Visual Genome 数据
Visual Genome 数据集是斯坦福大学维护的图像及图像内容语义信息的数据集，相比于著名的 ImageNet 图像标注数据集（也由斯坦福大学维护），Visual Genome 数据集附加了更为丰富的语义信息，用以拓展更加丰富的基于图像及语义信息的 人工智能 应用。目前包括 108249 张图片、420 万区域内容描述（Region Descriptions）、170 万图像内容问答（Visual Question Answers）、210 万对象案例（Object Instances）、180 万属性（Attributes）、180 万关系（Relationships）。该数据集最初于 2015 年由斯坦福大学发布，而后 2016 年发布 1.2 版本，2017 年发布 1.4 版本。
在关系识别方向，目标的类别为200，关系的类别为100.
1. Images
-
File image part1, image part2
全部 jpg 格式的图片
2. Image meta data
-
File image_data.json.zip
-
全部图片的 meta data，格式：
7. Relationships
-
全部的 relationships.