PascalVOC¶
- class src.dataset.pascal_voc.PascalVOC(sets, obj_resize)[source]¶
- filter_list()[source]¶
Filter out ‘truncated’, ‘occluded’ and ‘difficult’ images following the practice of previous works. In addition, this dataset has uncleaned label (in person category). They are omitted as suggested by README.
- get_multi(cls=None, num=2, shuffle=True, filter_outlier=True)[source]¶
Randomly get multiple objects from VOC-Berkeley keypoints dataset for multi-matching. The first image is fetched with all appeared keypoints, and the rest images are fetched with only inliers. :param cls: None for random class, or specify for a certain set :param num: number of objects to be fetched :param shuffle: random shuffle the keypoints :param filter_outlier: filter out outlier keypoints among images :return: (list of data, list of permutation matrices)
- get_pair(cls=None, shuffle=True, tgt_outlier=False, src_outlier=False)[source]¶
Randomly get a pair of objects from VOC-Berkeley keypoints dataset :param cls: None for random class, or specify for a certain set :param shuffle: random shuffle the keypoints :param src_outlier: allow outlier in the source graph (first graph) :param tgt_outlier: allow outlier in the target graph (second graph) :return: (pair of data, groundtruth permutation matrix)
- get_single_to_ref(idx, cls, shuffle=True)[source]¶
Get a single image, against a reference model containing all ground truth keypoints. :param idx: index in this class :param cls: specify for a certain class :param shuffle: random shuffle the keypoints :return: (data, groundtruth permutation matrix)
- property length¶