update

Author: ZPdesu

Panorama_Rectification/Batch_Simon_Panoramas_final.py

@@ -0,0 +1,222 @@
+import glob
+from default_params import default_params
+import skimage.io
+import numpy as np
+from PIL import Image, ImageDraw
+import matplotlib.pyplot as plt
+import os.path
+import skimage.io
+import time
+import json
+import multiprocessing
+
+import skimage.io
+from Panos.Pano_rectification import simon_rectification
+from Panos.Pano_project import project_face, stitch_tiles, render_imgs
+import matplotlib.pyplot as plt
+
+from Panos.Pano_visualization import R_heading, draw_all_vp_and_hl_color, draw_all_vp_and_hl_bi, \
+    draw_zenith_on_top_color, draw_zenith_on_top_bi, draw_sphere_zenith, R_roll, R_pitch
+from Panos.Pano_zp_hvp import calculate_consensus_zp
+from Panos.Pano_consensus_vis import draw_consensus_zp_hvps, draw_consensus_rectified_sphere, \
+    draw_center_hvps_rectified_sphere, draw_center_hvps_on_panorams
+import Pano_hvp
+from Panos.Pano_histogram import calculate_histogram
+from Panos.Pano_project import project_facade_for_refine
+
+########## add new random seed
+#np.random.seed(1)
+
+plot_redundant = False
+save_directly = True
+
+root = 'Pano_new'
+
+Country_city = 'New'
+
+new_count = 5
+
+tmp_count = str(new_count)
+
+
+
+Img_folder = os.path.join(root, Country_city, 'images/')
+inter_Dir = os.path.join(root, 'Pano_hl_z_vp/')
+
+
+
+imageList = glob.glob(Img_folder + '*.jpg')
+imageList.sort()
+
+rendering_output_folder = os.path.join(root, Country_city, 'Rendering')
+if not os.path.exists(rendering_output_folder):
+    os.makedirs(rendering_output_folder)
+
+
+#for im_path in ['/home/zhup/Desktop/GSV_Pano_val/Val/images/9wG3a9VOkwTSqnq6zsbdSQ.jpg']:
+# for im_path in imageList[10*new_count:10*(new_count+1)]:
+for im_path in imageList:
+    print(im_path)
+    im = Image.open(im_path)
+    # rendering_img_base = os.path.join(rendering_output_folder, os.path.splitext(os.path.basename(im_path))[0])
+    rendering_img_base = os.path.join(rendering_output_folder, os.path.splitext(os.path.basename(im_path))[0])
+
+    task = 'hahaha/'
+    thread_num = 1
+    thread = str(thread_num) + '/'
+    tmp_folder = os.path.join(root, Country_city, 'tmp', task, thread)
+
+    if not os.path.exists(tmp_folder):
+        os.makedirs(tmp_folder)
+    removelist = glob.glob(tmp_folder + '*.jpg')
+    for i in removelist:
+        os.remove(i)
+
+    #render_num = 16
+    render_num = 4
+    start = int(-render_num / 2) + 1
+    end = render_num + start
+    degree = 360 / render_num
+    panorama_img = skimage.io.imread(im_path)
+    coordinates_list = []
+
+
+    tilelist = render_imgs(panorama_img, tmp_folder, save_directly)
+    if not save_directly:
+        tilelist = glob.glob(tmp_folder + '*.jpg')
+        tilelist.sort()
+
+    hl = []
+    hvps = []
+    hvp_groups = []
+    z = []
+    z_group = []
+    ls = []
+    z_homo = []
+    hvp_homo = []
+    ls_homo = []
+
+    for i in range(len(tilelist)):
+        [tmp_hl, tmp_hvps, tmp_hvp_groups, tmp_z, tmp_z_group, tmp_ls, tmp_z_homo, tmp_hvp_homo, tmp_ls_homo, params] = simon_rectification(tilelist[i], i, inter_Dir, root, tmp_count)
+        hl.append(tmp_hl)
+        hvps.append(tmp_hvps)
+        hvp_groups.append(tmp_hvp_groups)
+        z.append(tmp_z)
+        z_group.append(tmp_z_group)
+        ls.append(tmp_ls)
+        z_homo.append(tmp_z_homo)
+        hvp_homo.append(tmp_hvp_homo)
+        ls_homo.append(tmp_ls_homo)
+
+    # print('get all the zenith points')
+
+
+    removelist = glob.glob(tmp_folder + '*.jpg')
+    for i in removelist:
+        os.remove(i)
+
+    ####################### Get all the zenith points from all the (8) viewpoints
+
+    zenith_points = np.array([R_heading(np.pi / 2 * (i - 1)).dot(zenith) for i, zenith in enumerate(z_homo)])
+    points2 = np.array([R_heading(np.pi / 2 * (i - 1)).dot(np.array([0., 0., 1.])) for i in range(len(z_homo))])
+    hv_points = [(R_heading(np.pi / 2 * (i - 1)).dot(hv_p.T)).T for i, hv_p in enumerate(hvp_homo)]
+
+
+    if plot_redundant:
+        draw_all_vp_and_hl_color(zenith_points, hv_points, im.copy(), root)
+
+        draw_all_vp_and_hl_bi(zenith_points, hv_points, im.copy(), root)
+        #draw_zenith_on_top_color(zenith_points, root)
+        #draw_zenith_on_top_bi(zenith_points, root)
+        draw_sphere_zenith(zenith_points, hv_points, root)
+
+
+
+    ####################### Calculate the consensus zenith point
+
+    [zenith_consensus, best_zenith] = calculate_consensus_zp(zenith_points, method='svd')
+
+    # Transform the zenith points back to original homogeneous coordinates
+    # zenith_consensus_org = np.array(
+    #     [R_heading(-np.pi / 6 * (i - 5)).dot(zenith) for i, zenith in enumerate(zenith_consensus)])
+    zenith_consensus_org = np.array([R_heading(-np.pi / 2 * (i - 1)).dot(zenith) for i, zenith in enumerate(zenith_consensus)])
+
+
+
+    result_list = []
+    for i in range(len(zenith_consensus_org)):
+
+        #result = Pano_hvp.hvp_from_zenith(ls_homo[i], zenith_consensus_org[i], params)
+
+        result = Pano_hvp.get_all_hvps(ls_homo[i], zenith_consensus_org[i], params)
+        result_list.append(result)
+
+    hvps_consensus_org = []
+    for i in range(len(result_list)):
+        # hvps_consensus_org.append(result_list[i]['hvp_homo'])
+        hvps_consensus_org.append(result_list[i])
+
+    hvps_consensus_uni = [(R_heading(np.pi / 2 * (i - 1)).dot(hv_p.T)).T for i, hv_p in enumerate(hvps_consensus_org)]
+
+    if plot_redundant:
+        draw_consensus_zp_hvps(best_zenith, hvps_consensus_uni, im.copy(), root)
+
+
+    ####################### Calculate pitch and roll
+    pitch = np.arctan(best_zenith[2] / best_zenith[1])
+    roll = - np.arctan(best_zenith[0] / np.sign(best_zenith[1]) * np.hypot(best_zenith[1], best_zenith[2]))
+
+
+    hvps_consensus_rectified = [R_roll(-roll).dot(R_pitch(-pitch).dot(vp.T)).T for vp in hvps_consensus_uni]
+
+    if plot_redundant:
+        draw_consensus_rectified_sphere(hvps_consensus_rectified, root)
+
+
+
+    ###################### Calculate horizontal VP histogram
+
+    final_hvps_rectified = calculate_histogram(hvps_consensus_rectified, root, plot_redundant)
+
+    # final_hvps_rectified = [np.array([1.,0.,0.]), np.array([0.,0.,1.])]
+    # pitch = 0
+    # roll = 0
+
+
+    # rrd = np.random.rand() * np.pi
+    # final_hvps_rectified = [np.array([np.sin(rrd), 0., np.cos(rrd)]), np.array([np.sin(np.pi/2 + rrd), 0., np.cos(np.pi/2 + rrd)])]
+    # pitch = 0
+    # roll = 0
+
+
+
+
+    # Test whether the main vanishing point is near 90 degrees to the second vanishing point
+
+    # if len(final_hvps_rectified) == 2:
+    #     hvp1 = final_hvps_rectified[0]
+    #     hvp2 = final_hvps_rectified[1]
+    #     if np.abs(hvp1.dot(hvp2)) > np.sin(np.radians(5)):
+    #         final_hvps_rectified = [final_hvps_rectified[0]]
+
+    if plot_redundant:
+        draw_center_hvps_rectified_sphere(np.array(final_hvps_rectified), root)
+        draw_center_hvps_on_panorams(best_zenith, np.array(final_hvps_rectified), im.copy(), pitch, roll, root)
+
+
+
+    # Draw rectified panorama
+
+    # from Panos.Pano_new_pano import create_new_panorama, draw_new_panorama
+    # if plot_redundant:
+    #     new_pano_path = create_new_panorama(im_path, pitch, roll, root)
+    #     draw_new_panorama(new_pano_path, np.array(final_hvps_rectified), root)
+
+
+
+    ###################### Rendering images from panoramas
+
+    project_facade_for_refine(np.array(final_hvps_rectified), im.copy(), pitch, roll, im_path, root, tmp_folder, rendering_img_base, tmp_count)
+    print(100)
+
+