图像空间关系数据集

图像示例：

代码和相关论文：

SpatialSense: An Adversarially Crowdsourced Benchmark for Spatial Relation Recognition

Kaiyu Yang, Olga Russakovsky, and Jia Deng International Conference on Computer Vision (ICCV) 2019

数据引用：

@inproceedings{yang2019spatialsense,
  title={SpatialSense: An Adversarially Crowdsourced Benchmark for Spatial Relation Recognition},
  author={Yang, Kaiyu and Russakovsky, Olga and Deng, Jia},
  booktitle={International Conference on Computer Vision},
  year={2019},
}

标注文件信息：

{
  'url': 'https://farm4.staticflickr.com/3543/5704634119_8b8ccf3229.jpg',  # URL for Flickr Images
  'nsid': '10392797@N03',                                                  # Flickr NSID of the user
  'height': 500,
  'width': 281,
  'split': 'train',                                                        # train/valid/test split
  'annotations': [{                                                        # a list of spatial relations
    '_id': '59fbffe4f25c8070bb77ec42',                                     # an unique identifier for the relation      
    'predicate': 'on',               
    'object': {
      'y': 402,                                                            # (x, y) is a point on the object
      'x': 148,
      'name': 'ground',
      'bbox': [196, 500, 3, 278]                                           # bounding box
    },
    'subject': {
      'y': 317, 
      'x': 157, 
      'name': 'cat', 
      'bbox': [230, 434, 31, 264]
    },
    'label': True                                                          # the relation is a positive example
    }, {
    '_id': '59ff0e910de0c80e4077c5f0',
    'predicate': 'to the left of',
    'object': {
      'y': 213,
      'x': 240,
      'name': 'mirror',
      'bbox': [0, 345, 160, 280]
    },
    'subject': {
      'y': 303, 
      'x': 143, 
      'name': 'cat', 
      'bbox': [226, 449, 33, 271]
    },
    'label': True},
  ...  
  ]}

坐标和边界框的约定：

原点是图像的左上角。x轴沿宽度，y轴仅沿高度。边界框[y0, y1, x0, x1]有(x0, y0)其左上角，并(x1, y1)为它的右下角。

更多信息请查看项目主页。