issue with get_images function

[jiaxx]

There was a problem using get_images() to generate images for the objectome models (but not others). The output images sometimes look like part of an image. Also the object is not always presented at the desired position of the image, if there is any.

Here is my code:

First, I define a class in dataset.py:

class GenerativeDataset_rotation(GenerativeDatasetBase):

models =["pear_obj_2","single_pineapple","lo_poly_animal_CHICKDEE","lo_poly_animal_DUCK","lo_poly_animal_ELE_AS1","lo_poly_animal_RHINO_2","lo_poly_animal_TRTL_B","lo_poly_animal_TRANTULA"]
bad_backgrounds = ['INTERIOR_13ST.jpg', 'INTERIOR_12ST.jpg',
'INTERIOR_11ST.jpg', 'INTERIOR_10ST.jpg',
'INTERIOR_09ST.jpg', 'INTERIOR_08ST.jpg',
'INTERIOR_07ST.jpg', 'INTERIOR_06ST.jpg',
'INTERIOR_05ST.jpg']
good_backgrounds = [_b for _b in model_info.BACKGROUNDS
if _b not in bad_backgrounds]
templates = [
{'n_ex_per_model': 1,
'name': 'rotation', 
'template': {'bgname': choice(good_backgrounds),
'bgscale': 1.,
'bgpsi': 0,
'bgphi': 40.,
's': 1,
'ty': 0,
'tz': 0,
'ryz': choice([-160, -70, 0, 70, 160]),
'rxy': choice([-160, -70, 0, 70, 160]),
'rxz': choice([-160, -70, 0, 70, 160]),
}
}
]

Use genthor to generat meta and then images

import genthor.datasets as gd

dataset = gd.GenerativeDataset_rotation()
meta=dataset.meta

preproc = {'dtype':'uint8', 'size':(1024, 1024), 'normalize':False, 'mode':'L'}
im = dataset.get_images(preproc, 
#global_light_spec=[{'name': 'bgalight', 'type': 'AmbientLight', 'color':(1,1,1,.01)}]
global_light_spec=None)

fig = plt.figure(figsize=(6, 6))
ax = plt.imshow(im[0][::-1], cmap='gray')