def image_encode(args, i, item, q_out):
oitem = [item.id]
#print('flag', item.flag)
if item.flag == 0:
fullpath = item.image_path
header = mx.recordio.IRHeader(item.flag, item.label, item.id, 0)
#print('write', item.flag, item.id, item.label)
if item.aligned:
with open(fullpath, 'rb') as fin:
img = fin.read()
s = mx.recordio.pack(header, img)
q_out.put((i, s, oitem))
else:
img = cv2.imread(fullpath, args.color)
assert item.landmark is not None
img = face_align.norm_crop(img, item.landmark)
s = mx.recordio.pack_img(header,
img,
quality=args.quality,
img_fmt=args.encoding)
q_out.put((i, s, oitem))
else:
header = mx.recordio.IRHeader(item.flag, item.label, item.id, 0)
#print('write', item.flag, item.id, item.label)
s = mx.recordio.pack(header, '')
q_out.put((i, s, oitem))
def crop_face(faces, imgName, imgMat):
savePath = './images'
imgName = imgName.split('/')[-1]
index = 0
print(imgMat.shape)
for num in range(faces.shape[0]):
kpoint = faces[num, 5:15]
kpoint = np.array(kpoint).reshape(5, 2)
cropImg = norm_crop(imgMat, kpoint)
cv2.imwrite(os.path.join(savePath, '{}_{}'.format(index, imgName)),
cropImg)
index += 1
def get_norm_crop(image_path):
im = cv2.imread(image_path)
im_shape = im.shape
im_size_min = np.min(im_shape[0:2])
im_size_max = np.max(im_shape[0:2])
im_scale = float(target_size) / float(im_size_min)
# prevent bigger axis from being more than max_size:
if np.round(im_scale * im_size_max) > max_size:
im_scale = float(max_size) / float(im_size_max)
bbox, landmark = detector.detect(im, threshold=0.5, scales=[im_scale])
if bbox.shape[0] == 0:
bbox, landmark = detector.detect(
im,
threshold=0.05,
scales=[im_scale * 0.75, im_scale, im_scale * 2.0])
# print('refine', im.shape, bbox.shape, landmark.shape)
nrof_faces = bbox.shape[0]
if nrof_faces > 0:
det = bbox[:, 0:4]
img_size = np.asarray(im.shape)[0:2]
bindex = 0
if nrof_faces > 1:
bounding_box_size = (det[:, 2] - det[:, 0]) * (det[:, 3] -
det[:, 1])
img_center = img_size / 2
offsets = np.vstack([(det[:, 0] + det[:, 2]) / 2 - img_center[1],
(det[:, 1] + det[:, 3]) / 2 - img_center[0]])
offset_dist_squared = np.sum(np.power(offsets, 2.0), 0)
bindex = np.argmax(bounding_box_size - offset_dist_squared *
2.0) # some extra weight on the centering
_bbox = det[bindex, 0:4]
top, bottom, left, right = adjust_bbox(_bbox, im_shape, 0.1)
cropped = im[top:bottom, left:right]
cropped = cv2.resize(cropped, (args.image_size, args.image_size))
_landmark = landmark[bindex]
warped = face_align.norm_crop(im,
landmark=_landmark,
image_size=args.image_size,
mode=args.align_mode)
return warped, cropped
else:
return None
def test(FAPrefix, SSHPrefix, imgPath):
FAPredictor = FaceAttributesPredictor(FAPrefix, epoch=1, ctxId=0)
faceDetector = SSHDetector(SSHPrefix, epoch=0, ctx_id=0)
images_name = get_images_name_list(imgPath)
for img in images_name:
print('image: %s' % (img))
imgMat = cv2.imread(img)
faces = faceDetector.detect(imgMat, threshold=0.8)
print(faces.shape[0], ' faces detected.\n')
imgMatList = []
for num in range(faces.shape[0]):
kpoint = faces[num, 5:15]
kpoint = np.array(kpoint).reshape(5, 2)
cropImg = norm_crop(imgMat, kpoint)
imgMatList.append(cropImg)
faResult = FAPredictor.predict(imgMatList)
assert len(faResult) == len(imgMatList)
show_face_rec_kpoint_attribute(faces, img, imgMat, faResult)
def get_align_input(self, img, points):
nimg = face_align.norm_crop(
img, landmark=points, image_size=112, mode='arcface')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2, 0, 1))
return aligned