:param src_img: 用于wrap的原图 :param corners_list: [tensor, tensor ...]，用于corner2bbox函数 :param out_size: 生成的透视后图像尺寸 w*h ?? :return:

:param src_img: 用于wrap的原图 :param corners_list: [tensor, tensor ...]，用于corner2bbox函数 :param out_size: 生成的透视后图像尺寸 w*h ?? :return:

:param index: :return: img_tensor size(3, 512, 512); corners_tensor size(N, 8); lp_label_tensor list(obj_num), ea

:param feature_infos: element 0 means sizes((32, 32), (16, 16), ...), element 1 means channels

:param chn_in: :param size_hw_in: (h, w) or [h, w] :param chn_shrink_ratio:

:param chn_in: a number or a list (tuple) :param size_hw_in: (h, w) or [h, w] :param chn_shrink_ratio:

:param chn_in: :param mode: choose in (0, 1, 2, 3), 0:skip concatenate connection; 1:skip concatenate & add connection;

:param feature_infos: element 0 means sizes((32, 32), (16, 16), ...), element 1 means channels

:param feature_infos: element 0 means sizes((32, 32), (16, 16), ...), element 1 means channels

:param chn_in: :param size_hw_in: (h, w) or [h, w] :param chn_shrink_ratio:

:param chn_in: a number or a list (tuple) :param size_hw_in: (h, w) or [h, w] :param chn_shrink_ratio:

:param chn_in: :param mode: choose in (0, 1, 2, 3), 0:skip concatenate connection; 1:skip concatenate & add connection;

返回image的tensor

:param single_reg_map: size(32, 32, 8) :param stage_lvl: scale ratio :param S0: 16 from fovea box :return:

:param single_reg_map: size(32, 32, 8) :param stage_lvl: scale ratio :param S0: 16 from fovea box :return:

:param corners_list: [tensor, tensor ...] :return: list of box, float, not tensor

:param corners_list: [tensor, tensor ...] :return: list of box, float, not tensor

:param corners_maps: tensor size(B, H, W, 8) :return: size(B, H, W, 4)

:param corners_maps: tensor size(B, H, W, 8) :return: size(B, H, W, 4)

:param corners: size(N, 8) :return: size(N, 4)

:param corners: size(N, 8) :return: size(N, 4)

:param corner: tensor size(8) :param effective_ratio: :param ignore_ratio: :param max_size :return: as input 'corner'

:param corner: tensor size(8) :param effective_ratio: :param ignore_ratio: :param max_size :return: as input 'corner'

:param img_path: full image path for evaluation :param save_path: to save the eval image :param out_corner: corner after nms of output

:param img_path: full image path for evaluation :param save_path: to save the eval image :param out_corner: corner after nms of output

:param predict_maps: outputs from detection network, size(B, 32, 32, 12) :param top_k: :param value_threshold: :param nms_thresho

:param predict_maps: outputs from detection network, size(B, 32, 32, 12) :param top_k: :param value_threshold: :param nms_thresho

:param corners_list: tensor(8), eight corner coordinate, clockwise :param point: tensor(N, 2), to be distinguished :return: bool tenso

:param corners_list: tensor(8), eight corner coordinate, clockwise :param point: tensor(N, 2), to be distinguished :return: bool tenso

:param fea_maps: feature maps to Perspective Transform, size(C, H, W) :param corners_tensor: size(N, 8), N corner pairs number

:param fea_maps: feature maps to Perspective Transform, size(C, H, W) :param corners_tensor: size(N, 8), N corner pairs number

:param corner_preds: from function: detection_analysis, tensor size(obj1, 8) :param corner_targets: tensor, size(obj2, 8) :param gauss

:param corner_preds: from function: detection_analysis, tensor size(obj1, 8) :param corner_targets: tensor, size(obj2, 8) :param gauss

:param out_centers: size(B, H, W, 2) :param target_centers: size(B, H, W, 2) :param out_bbox: size(B, H, W, 4) :p