OCCLUSION_RESULT_CAT = []
for target_category in ['car']: #categories:
plot_out_dir = plot_out_dir_level + target_category + '/'
if not os.path.exists(plot_out_dir):
os.makedirs(plot_out_dir)
with torch.no_grad():
OCCLUSION_RESULT_TYPE = []
for occ_type in occ_types:
# Load data
test_imgs, test_labels, test_masks = getImg(
'test',
categories_train,
dataset,
data_path, [target_category],
occ_level,
occ_type,
bool_load_occ_mask=True)
nsamples = np.floor(len(test_imgs) *
nsamples_ratio).astype(np.int)
test_imgs = test_imgs[:nsamples]
test_labels = test_labels[:nsamples]
test_masks = test_masks[:nsamples]
test_imgset = Imgset(test_imgs,
test_masks,
test_labels,
imgLoader,
bool_square_images=False)
test_loader = DataLoader(dataset=test_imgset,
batch_size=1,
val_masks = []
# get training and validation images
for occ_level in occ_levels_train:
if occ_level == 'ZERO':
occ_types = ['']
train_fac = 0.9
else:
occ_types = ['_white', '_noise', '_texture', '']
train_fac = 0.1
for occ_type in occ_types:
imgs, labels, masks = getImg('train',
categories_train,
dataset,
data_path,
categories,
occ_level,
occ_type,
bool_load_occ_mask=False)
nimgs = len(imgs)
for i in range(nimgs):
if (random.randint(0, nimgs - 1) / nimgs) <= train_fac:
train_imgs.append(imgs[i])
train_labels.append(labels[i])
train_masks.append(masks[i])
elif not bool_train_with_occluders:
val_imgs.append(imgs[i])
val_labels.append(labels[i])
val_masks.append(masks[i])
print('Total imgs for train ' + str(len(train_imgs)))