# -*- coding: utf-8 -*-
"""
File Name： gt_utils
Description : gt 四边形分割为固定宽度的系列gt boxes
Author : mick.yi
date： 2019/3/18
"""
import numpy as np
def linear_fit_y(xs, ys, x_list):
"""
线性函数拟合两点(x1,y1),(x2,y2)；并求得x_list在的取值
:param xs: [x1,x2]
:param ys: [y1,y2]
:param x_list: x轴坐标点,numpy数组 [n]
:return:
"""
if xs[0] == xs[1]: # 垂直线
return np.ones_like(x_list) * np.mean(ys)
elif ys[0] == ys[1]: # 水平线
return np.ones_like(x_list) * ys[0]
else:
fn = np.poly1d(np.polyfit(xs, ys, 1)) # 一元线性函数
return fn(x_list)
def get_min_max_y(quadrilateral, xs):
"""
获取指定x值坐标点集合四边形上的y轴最小值和最大值
:param quadrilateral: 四边形坐标；x1,y1,x2,y2,x3,y3,x4,y4
:param xs: x轴坐标点,numpy数组 [n]
:return: x轴坐标点在四边形上的最小值和最大值
"""
x1, y1, x2, y2, x3, y3, x4, y4 = quadrilateral.tolist()
y_val_1 = linear_fit_y(np.array([x1, x2]), np.array([y1, y2]), xs)
y_val_2 = linear_fit_y(np.array([x2, x3]), np.array([y2, y3]), xs)
y_val_3 = linear_fit_y(np.array([x3, x4]), np.array([y3, y4]), xs)
y_val_4 = linear_fit_y(np.array([x4, x1]), np.array([y4, y1]), xs)
y_val_min = []
y_val_max = []
for i in range(len(xs)):
y_val = []
if min(x1, x2) <= xs[i] <= max(x1, x2):
y_val.append(y_val_1[i])
if min(x2, x3) <= xs[i] <= max(x2, x3):
y_val.append(y_val_2[i])
if min(x3, x4) <= xs[i] <= max(x3, x4):
y_val.append(y_val_3[i])
if min(x4, x1) <= xs[i] <= max(x4, x1):
y_val.append(y_val_4[i])
# print("y_val:{}".format(y_val))
y_val_min.append(min(y_val))
y_val_max.append(max(y_val))
return np.array(y_val_min), np.array(y_val_max)
def get_xs_in_range(x_array, x_min, x_max):
"""
获取分割坐标点
:param x_array: 宽度方向分割坐标点数组；0~image_width,间隔16 ；如:[0,16,32,...608]
:param x_min: 四边形x最小值
:param x_max: 四边形x最大值
:return:
"""
indices = np.logical_and(x_array >= x_min, x_array <= x_max)
xs = x_array[indices]
# 处理两端的值
if xs.shape[0] == 0 or xs[0] > x_min:
xs = np.insert(xs, 0, x_min)
if xs.shape[0] == 0 or xs[-1] < x_max:
xs = np.append(xs, x_max)
return xs
def gen_gt_from_quadrilaterals(gt_quadrilaterals, input_gt_class_ids, image_shape, width_stride, box_min_size=3):
"""
从gt 四边形生成，宽度固定的gt boxes
:param gt_quadrilaterals: GT四边形坐标,[n,(x1,y1,x2,y2,x3,y3,x4,y4)]
:param input_gt_class_ids: GT四边形类别，一般就是1 [n]
:param image_shape:
:param width_stride: 分割的步长，一般16
:param box_min_size: 分割后GT boxes的最小尺寸
:return:
gt_boxes：[m,(y1,x1,y2,x2)]
gt_class_ids: [m]
"""
h, w = list(image_shape)[:2]
x_array = np.arange(0, w + 1, width_stride, np.float32) # 固定宽度间隔的x坐标点
# 每个四边形x 最小值和最大值
x_min_np = np.min(gt_quadrilaterals[:, ::2], axis=1)
x_max_np = np.max(gt_quadrilaterals[:, ::2], axis=1)
gt_boxes = []
gt_class_ids = []
for i in np.arange(len(gt_quadrilaterals)):
xs = get_xs_in_range(x_array, x_min_np[i], x_max_np[i]) # 获取四边形内的x中坐标点
ys_min, ys_max = get_min_max_y(gt_quadrilaterals[i], xs)
# print("xs:{}".format(xs))
# 为每个四边形生成固定宽度的gt
for j in range(len(xs) - 1):
x1, x2 = xs[j], xs[j + 1]
y1, y2 = np.min(ys_min[j:j + 2]), np.max(ys_max[j:j + 2])
gt_boxes.append([y1, x1, y2, x2])
gt_class_ids.append(input_gt_class_ids[i])
gt_boxes = np.reshape(np.array(gt_boxes), (-1, 4))
gt_class_ids = np.reshape(np.array(gt_class_ids), (-1,))
# 过滤高度太小的边框
height = gt_boxes[:, 2] - gt_boxes[:, 0]
width = gt_boxes[:, 3] - gt_boxes[:, 1]
indices = np.where(np.logical_and(height >= 8, width >= 2))
return gt_boxes[indices], gt_class_ids[indices]
