| import numpy as np |
| import cv2 |
| import matplotlib.pyplot as plt |
| import matplotlib.patches as patches |
| from lib.utils.img_utils import colors |
| from lib.utils import img_utils |
| from lib.utils.snake import snake_cityscapes_utils, snake_config |
|
|
|
|
| R = 8 |
| GREEN = (18, 127, 15) |
| WHITE = (255, 255, 255) |
|
|
|
|
| def visualize_snake_detection(img, data): |
|
|
| def blend_hm_img(hm, img): |
| hm = np.max(hm, axis=0) |
| h, w = hm.shape[:2] |
| img = cv2.resize(img, dsize=(w, h), interpolation=cv2.INTER_LINEAR) |
| hm = np.array([255, 255, 255]) - (hm.reshape(h, w, 1) * colors[0]).astype(np.uint8) |
| ratio = 0.5 |
| blend = (img * ratio + hm * (1 - ratio)).astype(np.uint8) |
| return blend |
|
|
| img = img_utils.bgr_to_rgb(img) |
| blend = blend_hm_img(data['ct_hm'], img) |
|
|
| plt.imshow(blend) |
| ct_ind = np.array(data['ct_ind']) |
| w = img.shape[1] // snake_config.down_ratio |
| xs = ct_ind % w |
| ys = ct_ind // w |
| for i in range(len(data['wh'])): |
| w, h = data['wh'][i] |
| x_min, y_min = xs[i] - w / 2, ys[i] - h / 2 |
| x_max, y_max = xs[i] + w / 2, ys[i] + h / 2 |
| plt.plot([x_min, x_min, x_max, x_max, x_min], [y_min, y_max, y_max, y_min, y_min]) |
| plt.show() |
|
|
|
|
| def visualize_cp_detection(img, data): |
| act_ind = data['act_ind'] |
| awh = data['awh'] |
|
|
| act_hm_w = data['act_hm'].shape[2] |
| cp_h, cp_w = data['cp_hm'][0].shape[1], data['cp_hm'][0].shape[2] |
|
|
| img = img_utils.bgr_to_rgb(img) |
| plt.imshow(img) |
|
|
| for i in range(len(act_ind)): |
| act_ind_ = act_ind[i] |
| ct = act_ind_ % act_hm_w, act_ind_ // act_hm_w |
| w, h = awh[i] |
| abox = np.array([ct[0] - w/2, ct[1] - h/2, ct[0] + w/2, ct[1] + h/2]) |
|
|
| cp_ind_ = data['cp_ind'][i] |
| cp_wh_ = data['cp_wh'][i] |
|
|
| for j in range(len(cp_ind_)): |
| ct = cp_ind_[j] % cp_w, cp_ind_[j] // cp_w |
| x = ct[0] / cp_w * w |
| y = ct[1] / cp_h * h |
| x_min = (x - cp_wh_[j][0] / 2 + abox[0]) * snake_config.down_ratio |
| y_min = (y - cp_wh_[j][1] / 2 + abox[1]) * snake_config.down_ratio |
| x_max = (x + cp_wh_[j][0] / 2 + abox[0]) * snake_config.down_ratio |
| y_max = (y + cp_wh_[j][1] / 2 + abox[1]) * snake_config.down_ratio |
| plt.plot([x_min, x_min, x_max, x_max, x_min], [y_min, y_max, y_max, y_min, y_min]) |
|
|
| plt.show() |
|
|
|
|
| def visualize_snake_evolution(img, data): |
| img = img_utils.bgr_to_rgb(img) |
| plt.imshow(img) |
| for poly in data['i_gt_py']: |
| poly = poly * 4 |
| poly = np.append(poly, [poly[0]], axis=0) |
| plt.plot(poly[:, 0], poly[:, 1]) |
| plt.scatter(poly[0, 0], poly[0, 1], edgecolors='w') |
| plt.show() |
|
|
|
|
| def visualize_snake_octagon(img, extreme_points): |
| img = img_utils.bgr_to_rgb(img) |
| octagons = [] |
| bboxes = [] |
| ex_points = [] |
| for i in range(len(extreme_points)): |
| for j in range(len(extreme_points[i])): |
| bbox = get_bbox(extreme_points[i][j]*4) |
| octagon = snake_cityscapes_utils.get_octagon(extreme_points[i][j]*4) |
| bboxes.append(bbox) |
| octagons.append(octagon) |
| ex_points.append(extreme_points[i][j]) |
| _, ax = plt.subplots(1) |
| ax.imshow(img) |
| n = len(octagons) |
| for i in range(n): |
| x, y, x_max, y_max = bboxes[i] |
| ax.add_patch(patches.Polygon(xy=[[x, y], [x, y_max], [x_max, y_max], [x_max, y]], fill=False, linewidth=1, |
| edgecolor='r')) |
| octagon = np.append(octagons[i], octagons[i][0]).reshape(-1, 2) |
| ax.plot(octagon[:, 0], octagon[:, 1]) |
| ax.scatter(ex_points[i][:, 0] * 4, ex_points[i][:, 1] * 4, edgecolors='w') |
| plt.show() |
|
|
|
|
| def get_bbox(ex): |
| x = ex[:, 0] |
| y = ex[:, 1] |
| bbox = [np.min(x), np.min(y), np.max(x), np.max(y)] |
| return bbox |
|
|
