Open Images V4

Open Images V4

Open Images 包含大约9百万张图片,带有 image-level 的标签和 object-level 的矩形框标注信息。

V4 的训练集包含174万张图片、1460万个矩形框和600个类别,是世界上最大的带有目标位置标注的数据集。矩形框由专业的标注人员标注,以确保准确性和一致性。而且图片种类多种多样、场景复杂,平均每张图片包含8.4个物体。同时, image-level 的标注分类为数千,详情如下表:

image-level 标注信息:

Train Validation Test # Classes # Trainable Classes
Images 9,011,219 41,620 125,436 - -
Machine-Generated Labels 78,977,695 512,093 1,545,835 7,870 4,764
Human-Verified Labels 27,894,289 pos: 13,444,569 neg: 14,449,720 551,390 pos: 365,772 neg: 185,618 1,667,399 pos: 1,105,052 neg: 562,347 19,794 7,186

object-level 标注信息:

Train Validation Test # Classes
Images 1,743,042 41,620 125,436 -
Boxes 14,610,229 204,621 625,282 600

更多详细信息可以参考:https://storage.googleapis.com/openimages/web/factsfigures.html

Object level

每一行定义一个矩形框:

1
2
3
4
5
6
7
8
9
10
CopyImageID,Source,LabelName,Confidence,XMin,XMax,YMin,YMax,IsOccluded,IsTruncated,IsGroupOf,IsDepiction,IsInside
000026e7ee790996,freeform,/m/07j7r,1,0.071905,0.145346,0.206591,0.391306,0,1,1,0,0
000026e7ee790996,freeform,/m/07j7r,1,0.439756,0.572466,0.264153,0.435122,0,1,1,0,0
000026e7ee790996,freeform,/m/07j7r,1,0.668455,1.000000,0.000000,0.552825,0,1,1,0,0
000062a39995e348,freeform,/m/015p6,1,0.205719,0.849912,0.154144,1.000000,0,0,0,0,0
000062a39995e348,freeform,/m/05s2s,1,0.137133,0.377634,0.000000,0.884185,1,1,0,0,0
0000c64e1253d68f,freeform,/m/07yv9,1,0.000000,0.973850,0.000000,0.043342,0,1,1,0,0
0000c64e1253d68f,freeform,/m/0k4j,1,0.000000,0.513534,0.321356,0.689661,0,1,0,0,0
0000c64e1253d68f,freeform,/m/0k4j,1,0.016515,0.268228,0.299368,0.462906,1,0,0,0,0
0000c64e1253d68f,freeform,/m/0k4j,1,0.481498,0.904376,0.232029,0.489017,1,0,0,0,0
  1. ImageID:图片 ID
  2. Source:标注来源。
    • freeformxclick 为手工标注。
    • activemil 使用方法1标注,并且人工验证保证 IoU>0.7。
  3. LabelName:标签ID。
  4. Confidence:置信度,总是为1。
  5. **XMin, XMax, YMin, YMax**:矩形框坐标,已经归一化,范围为[0,1]。
  6. IsOccluded: 物体是否被其他物体遮挡。
  7. IsTruncated: 物体是否被截断(超出图像范围)。
  8. IsGroupOf: 物体是否属于一组。 (如一群人),包含5个以上物体,且彼此接触、互相遮挡。
  9. IsDepiction: 物体是否为虚拟对象,如卡通人,不是现实世界物体。
  10. IsInside: 图片是否从物体内部拍摄, 如从汽车或者建筑内部拍摄。

Image level

1
2
3
4
5
6
7
8
9
10
CopyImageID,Source,LabelName,Confidence
000026e7ee790996,verification,/m/04hgtk,0
000026e7ee790996,verification,/m/07j7r,1
000026e7ee790996,crowdsource-verification,/m/01bqvp,1
000026e7ee790996,crowdsource-verification,/m/0csby,1
000026e7ee790996,verification,/m/01_m7,0
000026e7ee790996,verification,/m/01cbzq,1
000026e7ee790996,verification,/m/01czv3,0
000026e7ee790996,verification,/m/01v4jb,0
000026e7ee790996,verification,/m/03d1rd,0
  1. Source: 表明标签如何生成:
    • verification Google 标注人员验证。
    • crowdsource-verification Crowdsource App 验证。
    • machine 机器生成的标签。
  2. Confidence:人工验证的正样本为1,负样本为0,机器生成的为分数,通常 >= 0.5,其值越大,越不可能为 Fase Positive。
  3. Class Names:标签ID。
  4. ImageID:图片ID。

下载

可以再在这里找到详细的下载列表,如下图:

下载列表

而图片可以从Figure EightCVDF下载:

参考链接

  1. https://storage.googleapis.com/openimages/web/factsfigures.html
  2. https://ai.googleblog.com/2018/04/announcing-open-images-v4-and-eccv-2018.html

  1. We don’t need no bounding-boxes: Training object class detectors using only human verification, Papadopolous et al., CVPR 2016.

来源:

https://xblog.lufficc.com/blog/open-images-v4