def computeDPIoU(self, imgId, catId):
p = self.params
if p.useCats:
gt = self._gts[imgId, catId]
dt = self._dts[imgId, catId]
else:
gt = [_ for cId in p.catIds for _ in self._gts[imgId, cId]]
dt = [_ for cId in p.catIds for _ in self._dts[imgId, cId]]
if len(gt) == 0 and len(dt) == 0:
return []
inds = np.argsort([-d["score"] for d in dt], kind="mergesort")
dt = [dt[i] for i in inds]
if len(dt) > p.maxDets[-1]:
dt = dt[0 : p.maxDets[-1]]
gtmasks = []
for g in gt:
if DensePoseDataRelative.S_KEY in g:
# convert DensePose mask to a binary mask
mask = np.minimum(self.getDensePoseMask(g[DensePoseDataRelative.S_KEY]), 1.0)
_, _, w, h = g["bbox"]
scale_x = float(max(w, 1)) / mask.shape[1]
scale_y = float(max(h, 1)) / mask.shape[0]
mask = spzoom(mask, (scale_y, scale_x), order=1, prefilter=False)
mask = np.array(mask > 0.5, dtype=np.uint8)
rle_mask = self._generate_rlemask_on_image(mask, imgId, g)
elif "segmentation" in g:
segmentation = g["segmentation"]
if isinstance(segmentation, list) and segmentation:
# polygons
im_h, im_w = self.size_mapping[imgId]
rles = maskUtils.frPyObjects(segmentation, im_h, im_w)
rle_mask = maskUtils.merge(rles)
elif isinstance(segmentation, dict):
if isinstance(segmentation["counts"], list):
# uncompressed RLE
im_h, im_w = self.size_mapping[imgId]
rle_mask = maskUtils.frPyObjects(segmentation, im_h, im_w)
else:
# compressed RLE
rle_mask = segmentation
else:
rle_mask = self._generate_rlemask_on_image(None, imgId, g)
else:
rle_mask = self._generate_rlemask_on_image(None, imgId, g)
gtmasks.append(rle_mask)
dtmasks = []
for d in dt:
mask = self._extract_mask(d)
mask = np.require(np.asarray(mask > 0), dtype=np.uint8, requirements=["F"])
rle_mask = self._generate_rlemask_on_image(mask, imgId, d)
dtmasks.append(rle_mask)
# compute iou between each dt and gt region
iscrowd = [int(o["iscrowd"]) for o in gt]
iousDP = maskUtils.iou(dtmasks, gtmasks, iscrowd)
return iousDP
ct = nibabel.load(c1)
seg = nibabel.load(c2)
T = ct.get_sform()
ct = ct.get_data()
seg = seg.get_data()
res = [1.5, 1.5, 1.5]
zoom_factor = []
for k in range(3):
zoom_factor.append(abs(T[k, k]) / res[k])
zoom_factor = tuple(zoom_factor)
ct = np.round( spzoom( ct.astype(float), zoom_factor, order=1 ) ).astype('int16')
seg = np.round( spzoom( seg.astype(float), zoom_factor, order=0 ) ).astype('uint8')
print ('Size after zoom: ', ct.shape)
if ct.shape[0] < W or ct.shape[1] < W or ct.shape[2] < W:
print ("Patch size too big. Skipping ...")
small_files.append(f)
continue
'''
if ct.shape[0] < W:
p = W-ct.shape[0]
ct = np.pad(ct,( (p,0), (0,0), (0,0) ), mode='minimum')
seg = np.pad(seg,( (p,0), (0,0), (0,0) ), mode='minimum')
T = ct.affine
ct = ct.get_data()
seg = seg.get_data()
res = [1., 1., 1.]
W=96
zoom_factor = []
for k in range(3):
zoom_factor.append(abs(T[k, k]) / res[k])
zoom_factor = tuple(zoom_factor)
ct = np.round( spzoom( ct.astype(float), zoom_factor, order=1 ) ).astype('int16')
seg = np.round( spzoom( seg.astype(float), zoom_factor, order=0 ) ).astype('uint8')
xmax = ct.shape[0] - W
ymax = ct.shape[1] - W
zmax = ct.shape[2] - W
stride = W // 4
# Calculating indices at which to extract patches
z_list = list(range(0, zmax, stride))
z_list = [z for z in z_list if (z + W) < (ct.shape[2] - 1)]
if z_list:
z_list[-1] = zmax