Repeated initialization for "binay_masks" in inference.py

[ghost-ttt5]

In inference.py, "binay_masks" should be a list used to store binary masks.
It is firstly initialized when reading the binary masks from the json file.

InstanceDiffusion/inference.py

Lines 194 to 220 in 7306022

	binay_masks = []
	# class_names = []
	instance_captions = []
	points_list = []
	scribbles_list = []
	prompt = data['caption']
	crop_mask_image = False
	for inst_idx in range(len(data['annos'])):
	if "mask" not in data['annos'][inst_idx] or data['annos'][inst_idx]['mask'] == []:
	instance_mask = np.zeros((512,512,1))
	else:
	instance_mask = decodeToBinaryMask(data['annos'][inst_idx]['mask'])
	if crop_mask_image:
	# crop the instance_mask to 512x512, centered at the center of the instance_mask image
	# get the center of the instance_mask
	center = np.array([instance_mask.shape[0]//2, instance_mask.shape[1]//2])
	# get the top left corner of the crop
	top_left = center - np.array([256, 256])
	# get the bottom right corner of the crop
	bottom_right = center + np.array([256, 256])
	# crop the instance_mask
	instance_mask = instance_mask[top_left[0]:bottom_right[0], top_left[1]:bottom_right[1]]
	binay_masks.append(instance_mask)
	data['width'] = 512
	data['height'] = 512
	else:
	binay_masks.append(instance_mask)

But in the process of handling the missing binary masks, the "binay_masks" is initialized again. This would appear to cause previously read masks to be discarded, and subsequent steps that rely on binay_masks will use zero masks.

InstanceDiffusion/inference.py

Lines 248 to 251 in 7306022

	# get binary masks for each instance, if not provided, use all zeros
	binay_masks = []
	for i in range(len(locations) - len(binay_masks)):
	binay_masks.append(np.zeros((512,512,1)))