diff --git a/data/create_syn_data.py b/data/create_syn_data.py index c392a9e..809206e 100644 --- a/data/create_syn_data.py +++ b/data/create_syn_data.py @@ -11,6 +11,8 @@ import os import collections import sys +import xmltodict as xmltodict + sys.path.append('../') import renderer import co @@ -201,6 +203,16 @@ def create_data(out_root, idx, n_samples, imsize, patterns, K, baseline, blend_i print(f'create sample {idx + 1}/{n_samples} took {time.time() - tic}[s]') +def load_camera_parameters(): + with open('calibration_result.xml') as f: + cam_mat = xmltodict.parse(f.read()) + # weird casting cause the values are str(float) (eg. '123.'), but we want int + imsize = [int(float(x)) for x in cam_mat['opencv_storage']['img_shape']['data'].split()] + K_shape = int(cam_mat['opencv_storage']['cam_int']['rows']), int(cam_mat['opencv_storage']['cam_int']['cols']) + K = np.array(cam_mat['opencv_storage']['cam_int']['data'].split(), dtype=float).reshape(K_shape).T + return imsize, K + + if __name__ == '__main__': np.random.seed(42) @@ -230,11 +242,9 @@ if __name__ == '__main__': objs = get_objs(shapenet_root, obj_classes) # camera parameters - imsize = (488, 648) + imsize, K = load_camera_parameters() imsizes = [(imsize[0] // (2 ** s), imsize[1] // (2 ** s)) for s in range(4)] - # K = np.array([[567.6, 0, 324.7], [0, 570.2, 250.1], [0 ,0, 1]], dtype=np.float32) - K = np.array([[1929.5936336276382, 0, 113.66561071478046], [0, 1911.2517985448746, 473.70108079885887], [0, 0, 1]], - dtype=np.float32) + focal_lengths = [K[0, 0] / (2 ** s) for s in range(4)] baseline = 0.075 blend_im = 0.6