cvtools.data_augs.crop package

Module contents

class cvtools.data_augs.crop.CropLargeImages(dataset, crop_method, over_strict=True)[源代码]

基类：cvtools.data_augs.crop.crop_abc.Crop

crop_for_test()[源代码]
crop_for_train(over_samples=None)[源代码]

训练集裁剪

参数:over_samples (dict) -- {类别: 重采样次数， ...}
over_sample(img, anns, over_samples)[源代码]
save(to_file, limit_border=False)[源代码]
class cvtools.data_augs.crop.CocoDatasetForCrop(img_prefix, ann_file)[源代码]

基类：cvtools.data_augs.crop.crop_abc.CropDataset

recalc_anns(img_box, anns)[源代码]
save(crops, to_file, limit_border=False)[源代码]

通过自然索引对齐两组数据要小心

trans_ann(ann, img_box)[源代码]
class cvtools.data_augs.crop.CropImageInOrder(crop_w=1024, crop_h=1024, overlap=0.1, iof_th=0.7, size_th=1024)[源代码]

基类：cvtools.data_augs.crop.crop_abc.CropMethod

crop(img, anns=None)[源代码]
class cvtools.data_augs.crop.CropImageAdaptive(overlap=0.1, iof_th=0.7, small_prop=0.5, max_objs=100, slide_size=800, size_th=1024, strict_size=None)[源代码]

基类：cvtools.data_augs.crop.crop_abc.CropMethod

crop(img, anns=None)[源代码]

可能crop算法可能需要根据标签信息

class cvtools.data_augs.crop.CropImageProtected(iof_th=0.7, size_th=1024, strict=True)[源代码]

基类：cvtools.data_augs.crop.crop_abc.CropMethod

crop(img, anns=None)[源代码]