def segm_to_rle(segm, w, h):
"""
Convert segmentation map which can be polygons, uncompressed RLE to RLE.
Reference:
https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocotools/coco.py
Args:
segm (list<list<int>> or list<int>):
A segmentation map which can be polygons.
w (int): Image width
h (int): Image hight
Returns:
rle: RLE
"""
if type(segm) == list:
# polygon -- a single object might consist of multiple parts
# we merge all parts into one mask rle code
rles = maskUtils.frPyObjects(segm, h, w)
rle = maskUtils.merge(rles)
elif type(segm['counts']) == list:
# uncompressed RLE
rle = maskUtils.frPyObjects(segm, h, w)
else:
# rle
rle = segm
return rle
def poly2mask_single(h, w, poly):
# TODO: write test for poly2mask, using mask2poly convert mask to poly', compare poly with poly'
# visualize the mask
rles = maskUtils.frPyObjects(poly, h, w)
rle = maskUtils.merge(rles)
mask = maskUtils.decode(rle)
return mask
def annToRLE(ann, height, width):
"""
Convert annotation which can be polygons, uncompressed RLE to RLE.
:return: binary mask (numpy 2D array)
"""
segm = ann['segmentation']
if isinstance(segm, list):
# polygon -- a single object might consist of multiple parts
# we merge all parts into one mask rle code
rles = maskUtils.frPyObjects(segm, height, width)
rle = maskUtils.merge(rles)
elif isinstance(segm['counts'], list):
# uncompressed RLE
rle = maskUtils.frPyObjects(segm, height, width)
else:
# rle
rle = ann['segmentation']
return rle
def annToRLE(self, ann):
"""
Convert annotation which can be polygons, uncompressed RLE to RLE.
:return: binary mask (numpy 2D array)
"""
im = self.images.xs(ann.image_id)
h, w = im.height, im.width
segm = ann.segmentation
if type(segm) == list:
# polygon -- a single object might consist of multiple parts
# we merge all parts into one mask rle code
rles = maskUtils.frPyObjects(segm, h, w)
rle = maskUtils.merge(rles)
elif type(segm['counts']) == list:
# uncompressed RLE
rle = maskUtils.frPyObjects(segm, h, w)
else:
# rle
rle = ann['segmentation']
return rle
def poly2mask_single(h, w, poly):
# TODO: write test for poly2mask, using mask2poly convert mask to poly', compare poly with poly'
# visualize the mask
rles = maskUtils.frPyObjects(poly, h, w)
rle = maskUtils.merge(rles)
mask = maskUtils.decode(rle)
# sum = mask.sum()
# print("{} {} {} {}".format(sum, h, w, poly))
# if not mask.any():
# pass
return mask
def __getitem__(self, index):
img_file = os.path.join(self._img_dir,
str(self._infos.data[index].id) + '.jpg')
img = transforms.ToTensor()(Image.open(img_file).convert('RGB'))
img_w, img_h = img.size(2), img.size(1)
target = torch.LongTensor(img_h, img_w).zero_()
for inst in self._infos.data[index].insts:
polys = []
# { bg, person, bicycle, car, motorcycle, truck, bus, train }
if self._n_class == 7:
if inst.category_idx == 8:
inst.category_idx = 5
if inst.category_idx <= 6:
for poly in inst.seg:
polys.append(poly.tolist())
# { bg, person }
elif self._n_class == 2:
if inst.category_idx == 1:
for poly in inst.seg:
polys.append(poly.tolist())
if polys:
rles = maskUtils.frPyObjects(polys, img_h, img_w)
rle = maskUtils.merge(rles)
mask = maskUtils.decode(rle)
target.masked_fill_(torch.from_numpy(mask),
inst.category_idx)
p_w = self._infos.patchSize.w
p_h = self._infos.patchSize.h
x0 = random.randint(0, (img_w - p_w))
y0 = random.randint(0, (img_h - p_h))
img = img[:, y0:y0+p_h, x0:x0+p_w]
target = target[y0:y0+p_h, x0:x0+p_w]
return img, target
def main():
inputfile = '/home/qinjian/Segmentation/地理遥感图像分割/aicrowd房屋分割竞赛/val/images'
jsonfile = '/home/qinjian/Segmentation/地理遥感图像分割/aicrowd房屋分割竞赛/val/annotation-small.json'
outputfile = '/home/qinjian/Segmentation/地理遥感图像分割/aicrowd房屋分割竞赛/val/show'
mkdir_os(outputfile)
coco = COCO(jsonfile)
catIds = coco.getCatIds(catNms=['wires']) # catIds=1 表示人这一类
imgIds = coco.getImgIds(catIds=catIds) # 图片id，许多值
for i in range(len(imgIds)):
if i % 100 == 0:
print(i, "/", len(imgIds))
img = coco.loadImgs(imgIds[i])[0]
cvImage = cv2.imread(os.path.join(inputfile, img['file_name']), -1)
cvImage = cv2.cvtColor(cvImage, cv2.COLOR_BGR2GRAY)
cvImage = cv2.cvtColor(cvImage, cv2.COLOR_GRAY2BGR)
annIds = coco.getAnnIds(imgIds=img['id'], catIds=catIds, iscrowd=None)
anns = coco.loadAnns(annIds)
polygons = []
color = []
for ann in anns:
if 'segmentation' in ann:
if type(ann['segmentation']) == list:
# polygon
for seg in ann['segmentation']:
poly = np.array(seg).reshape((int(len(seg) / 2), 2))
poly_list = poly.tolist()
polygons.append(poly_list)
if ann['iscrowd'] == 0:
color.append([0, 0, 255])
if ann['iscrowd'] == 1:
color.append([0, 255, 255])
else:
exit()
print("-------------")
# mask
t = imgIds[ann['image_id']]
if type(ann['segmentation']['counts']) == list:
rle = maskUtils.frPyObjects([ann['segmentation']],
t['height'], t['width'])
else:
rle = [ann['segmentation']]
m = maskUtils.decode(rle)
if ann['iscrowd'] == 0:
color_mask = np.array([0, 0, 255])
if ann['iscrowd'] == 1:
color_mask = np.array([0, 255, 255])
mask = m.astype(np.bool)
cvImage[mask] = cvImage[mask] * 0.7 + color_mask * 0.3
point_size = 2
thickness = 2
for key in range(len(polygons)):
ndata = polygons[key]
cur_color = color[key]
for k in range(len(ndata)):
data = ndata[k]
cv2.circle(cvImage, (int(data[0]), int(data[1])), point_size,
(cur_color[0], cur_color[1], cur_color[2]),
thickness)
cv2.imwrite(os.path.join(outputfile, img['file_name']), cvImage)