def segments_pixeldist(self):
segmdist_fn = os.path.join(self.ds.path, 'cpmc', 'MySegmentsMat', self.name, 'top_masks_dists.mat')
if os.path.exists(segmdist_fn):
data = ml.loadmat(segmdist_fn)
dist = data['top_masks_dist']
segm_ids = data["segment_ids"]
else:
prop = self.top_masks
gt = self.ground_truth
dist, segm_ids = segments.get_class_distributions(gt, prop, num_classes=self.ds.classnum)
ml.savemat(segmdist_fn, {'top_masks_dist':dist, 'segment_ids': segm_ids})
segm_ids = list(np.array(segm_ids).ravel())
return dist, segm_ids
def segments_pixeldist(self):
segmdist_fn = os.path.join(self.ds.path, "cpmc", "MySegmentsMat", self.name, "top_masks_dists.mat")
if os.path.exists(segmdist_fn):
data = ml.loadmat(segmdist_fn)
dist = data["top_masks_dist"]
segm_ids = data["segment_ids"]
else:
prop = self.top_masks
gt = self.ground_truth
dist, segm_ids = segments.get_class_distributions(gt, prop, num_classes=self.ds.classnum)
ml.savemat(segmdist_fn, {"top_masks_dist": dist, "segment_ids": segm_ids})
segm_ids = list(np.array(segm_ids).ravel())
return dist, segm_ids
def object_masks(self):
ob_fn = os.path.join(self.ds.path, 'cpmc', 'MySegmentsMat', self.name, 'obj_gt_masks.mat')
if os.path.exists(ob_fn):
data = ml.loadmat(ob_fn)
masks = data['masks']
classes = data["classes"]
else:
masks = segments.seperate_gt_segments(self.object_ground_truth)
gt = self.ground_truth
dists, ids = segments.get_class_distributions(gt, masks)
classes = np.argmax(dists, axis=0)
assert masks.shape[2] == len(ids)
ml.savemat(ob_fn, {'masks':masks, 'classes': classes})
assert masks.shape[2] == classes.size
return masks, classes
def regions_pixeldist(self):
""" returns the regions distributions for the given image."""
regiondist_fn = os.path.join(self.ds.path, 'cpmc', 'MySegmentsMat', self.name, 'region_dists.mat')
if os.path.exists(regiondist_fn):
data = ml.loadmat(regiondist_fn)
dist = data['dists']
segm_ids = data["region_ids"]
else:
regions = self.regions
n_reg = np.max(regions) + 1
gt = self.ground_truth
prop = np.zeros((regions.shape[0], regions.shape[1], n_reg), dtype='bool')
for i in range(n_reg):
prop[:, :, i] = (regions == i)
dist, segm_ids = segments.get_class_distributions(gt, prop, num_classes=self.ds.classnum)
ml.savemat(regiondist_fn, {'dists':dist, 'region_ids': segm_ids})
segm_ids = list(np.array(segm_ids).ravel())
return dist, segm_ids
def object_masks(self):
ob_fn = os.path.join(self.ds.path, "cpmc", "MySegmentsMat", self.name, "obj_gt_masks.mat")
if os.path.exists(ob_fn):
data = ml.loadmat(ob_fn)
masks = data["masks"]
classes = data["classes"]
else:
masks = segments.seperate_gt_segments(self.object_ground_truth)
gt = self.ground_truth
dists, ids = segments.get_class_distributions(gt, masks)
classes = np.argmax(dists, axis=0)
assert masks.shape[2] == len(ids)
ml.savemat(ob_fn, {"masks": masks, "classes": classes})
assert masks.shape[2] == classes.size
return masks, classes
def regions_pixeldist(self):
""" returns the regions distributions for the given image."""
regiondist_fn = os.path.join(self.ds.path, "cpmc", "MySegmentsMat", self.name, "region_dists.mat")
if os.path.exists(regiondist_fn):
data = ml.loadmat(regiondist_fn)
dist = data["dists"]
segm_ids = data["region_ids"]
else:
regions = self.regions
n_reg = np.max(regions) + 1
gt = self.ground_truth
prop = np.zeros((regions.shape[0], regions.shape[1], n_reg), dtype="bool")
for i in range(n_reg):
prop[:, :, i] = regions == i
dist, segm_ids = segments.get_class_distributions(gt, prop, num_classes=self.ds.classnum)
ml.savemat(regiondist_fn, {"dists": dist, "region_ids": segm_ids})
segm_ids = list(np.array(segm_ids).ravel())
return dist, segm_ids