Merge branch 'experimental/align'
This commit is contained in:
Brandon Amos
commit
0e99d0c499
|
|
@ -13,6 +13,8 @@ images/examples-aligned
|
|||
|
||||
evaluation/*/*.csv
|
||||
evaluation/*/*.pdf
|
||||
evaluation/attic/*/*.csv
|
||||
evaluation/attic/*/*.pdf
|
||||
|
||||
demos/web/bower_components
|
||||
demos/web/unknown*.npy
|
||||
|
|
|
|||
|
|
@ -50,17 +50,19 @@ openfaceModelDir = os.path.join(modelDir, 'openface')
|
|||
|
||||
|
||||
def getRep(imgPath):
|
||||
img = cv2.imread(imgPath)
|
||||
if img is None:
|
||||
bgrImg = cv2.imread(imgPath)
|
||||
if bgrImg is None:
|
||||
raise Exception("Unable to load image: {}".format(imgPath))
|
||||
if args.verbose:
|
||||
print(" + Original size: {}".format(img.shape))
|
||||
rgbImg = cv2.cvtColor(bgrImg, cv2.COLOR_BGR2RGB)
|
||||
|
||||
bb = align.getLargestFaceBoundingBox(img)
|
||||
if args.verbose:
|
||||
print(" + Original size: {}".format(rgbImg.shape))
|
||||
|
||||
bb = align.getLargestFaceBoundingBox(rgbImg)
|
||||
if bb is None:
|
||||
raise Exception("Unable to find a face: {}".format(imgPath))
|
||||
|
||||
alignedFace = align.alignImg("affine", args.imgDim, img, bb)
|
||||
alignedFace = align.alignImg("affine", args.imgDim, bgrImg, bb)
|
||||
if alignedFace is None:
|
||||
raise Exception("Unable to align image: {}".format(imgPath))
|
||||
|
||||
|
|
@ -111,9 +113,6 @@ def infer(args):
|
|||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser()
|
||||
|
||||
parser.add_argument('--dlibFaceMean', type=str,
|
||||
help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "mean.csv"))
|
||||
parser.add_argument('--dlibFacePredictor', type=str,
|
||||
help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir,
|
||||
|
|
@ -158,11 +157,11 @@ network and classification models:
|
|||
|
||||
Use `--networkModel` to set a non-standard Torch network model.""")
|
||||
|
||||
sys.path.append(args.dlibRoot)
|
||||
sys.path = [args.dlibRoot] + sys.path
|
||||
import dlib
|
||||
from openface.alignment import NaiveDlib # Depends on dlib.
|
||||
|
||||
align = NaiveDlib(args.dlibFaceMean, args.dlibFacePredictor)
|
||||
align = NaiveDlib(args.dlibFacePredictor)
|
||||
net = openface.TorchWrap(
|
||||
args.networkModel, imgDim=args.imgDim, cuda=args.cuda)
|
||||
|
||||
|
|
|
|||
|
|
@ -44,8 +44,6 @@ openfaceModelDir = os.path.join(modelDir, 'openface')
|
|||
parser = argparse.ArgumentParser()
|
||||
|
||||
parser.add_argument('imgs', type=str, nargs='+', help="Input images.")
|
||||
parser.add_argument('--dlibFaceMean', type=str, help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "mean.csv"))
|
||||
parser.add_argument('--dlibFacePredictor', type=str, help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "shape_predictor_68_face_landmarks.dat"))
|
||||
parser.add_argument('--dlibRoot', type=str,
|
||||
|
|
@ -61,7 +59,7 @@ parser.add_argument('--verbose', action='store_true')
|
|||
|
||||
args = parser.parse_args()
|
||||
|
||||
sys.path.append(args.dlibRoot)
|
||||
sys.path = [args.dlibRoot] + sys.path
|
||||
import dlib
|
||||
|
||||
from openface.alignment import NaiveDlib # Depends on dlib.
|
||||
|
|
@ -70,7 +68,7 @@ if args.verbose:
|
|||
time.time() - start))
|
||||
|
||||
start = time.time()
|
||||
align = NaiveDlib(args.dlibFaceMean, args.dlibFacePredictor)
|
||||
align = NaiveDlib(args.dlibFacePredictor)
|
||||
net = openface.TorchWrap(args.networkModel, imgDim=args.imgDim, cuda=args.cuda)
|
||||
if args.verbose:
|
||||
print("Loading the dlib and OpenFace models took {} seconds.".format(
|
||||
|
|
@ -80,21 +78,23 @@ if args.verbose:
|
|||
def getRep(imgPath):
|
||||
if args.verbose:
|
||||
print("Processing {}.".format(imgPath))
|
||||
img = cv2.imread(imgPath)
|
||||
if img is None:
|
||||
bgrImg = cv2.imread(imgPath)
|
||||
if bgrImg is None:
|
||||
raise Exception("Unable to load image: {}".format(imgPath))
|
||||
rgbImg = cv2.cvtColor(bgrImg, cv2.COLOR_BGR2RGB)
|
||||
|
||||
if args.verbose:
|
||||
print(" + Original size: {}".format(img.shape))
|
||||
print(" + Original size: {}".format(rgbImg.shape))
|
||||
|
||||
start = time.time()
|
||||
bb = align.getLargestFaceBoundingBox(img)
|
||||
bb = align.getLargestFaceBoundingBox(rgbImg)
|
||||
if bb is None:
|
||||
raise Exception("Unable to find a face: {}".format(imgPath))
|
||||
if args.verbose:
|
||||
print(" + Face detection took {} seconds.".format(time.time() - start))
|
||||
|
||||
start = time.time()
|
||||
alignedFace = align.alignImg("affine", args.imgDim, img, bb)
|
||||
alignedFace = align.alignImg("affine", args.imgDim, rgbImg, bb)
|
||||
if alignedFace is None:
|
||||
raise Exception("Unable to align image: {}".format(imgPath))
|
||||
if args.verbose:
|
||||
|
|
@ -112,4 +112,4 @@ def getRep(imgPath):
|
|||
for (img1, img2) in itertools.combinations(args.imgs, 2):
|
||||
d = getRep(img1) - getRep(img2)
|
||||
print("Comparing {} with {}.".format(img1, img2))
|
||||
print(" + Squared l2 distance between representations: {}".format(np.dot(d, d)))
|
||||
print(" + Squared l2 distance between representations: {:0.3f}".format(np.dot(d, d)))
|
||||
|
|
|
|||
|
|
@ -25,8 +25,6 @@ import random
|
|||
|
||||
import cv2
|
||||
|
||||
from skimage import io
|
||||
|
||||
import openface
|
||||
from openface.alignment import NaiveDlib
|
||||
from openface.data import iterImgs
|
||||
|
|
|
|||
|
|
@ -54,8 +54,6 @@ dlibModelDir = os.path.join(modelDir, 'dlib')
|
|||
openfaceModelDir = os.path.join(modelDir, 'openface')
|
||||
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument('--dlibFaceMean', type=str, help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "mean.csv"))
|
||||
parser.add_argument('--dlibFacePredictor', type=str, help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "shape_predictor_68_face_landmarks.dat"))
|
||||
parser.add_argument('--dlibRoot', type=str,
|
||||
|
|
@ -72,11 +70,11 @@ parser.add_argument('--unknown', type=bool, default=False,
|
|||
|
||||
args = parser.parse_args()
|
||||
|
||||
sys.path.append(args.dlibRoot)
|
||||
sys.path = [args.dlibRoot] + sys.path
|
||||
import dlib
|
||||
from openface.alignment import NaiveDlib # Depends on dlib.
|
||||
|
||||
align = NaiveDlib(args.dlibFaceMean, args.dlibFacePredictor)
|
||||
align = NaiveDlib(args.dlibFacePredictor)
|
||||
net = openface.TorchWrap(args.networkModel, imgDim=args.imgDim, cuda=args.cuda)
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -5,8 +5,8 @@ on the standard
|
|||
[LFW](http://vis-www.cs.umass.edu/lfw/results.html)
|
||||
benchmark.
|
||||
We had to fallback to using the deep funneled versions for
|
||||
152 of 13233 images because dlib failed to detect a face or landmarks.
|
||||
We obtain a mean accuracy of 0.8483 ± 0.0172 with an AUC of 0.923.
|
||||
58 of 13233 images because dlib failed to detect a face or landmarks.
|
||||
We obtain a mean accuracy of 0.8138 ± 0.0149 with an AUC of 0.893.
|
||||
For comparison, training with Google-scale data results in an
|
||||
accuracy of .9963 ± 0.009.
|
||||
|
||||
|
|
|
|||
|
|
@ -13,18 +13,18 @@ Eric Clapton were generated with
|
|||
|---|---|---|---|
|
||||
| <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-1.jpg' width='200px'></img> | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-2.jpg' width='200px'></img> | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/clapton-1.jpg' width='200px'></img> | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/clapton-2.jpg' width='200px'></img> |
|
||||
|
||||
The following table shows that a distance threshold of `0.5` would
|
||||
The following table shows that a distance threshold of `0.8` would
|
||||
distinguish these two people.
|
||||
In practice, further experimentation should be done on the distance threshold.
|
||||
On our LFW experiments, the mean threshold across multiple
|
||||
experiments is 0.71 ± 0.027,
|
||||
experiments is about 0.68,
|
||||
see [accuracies.txt](https://github.com/cmusatyalab/openface/blob/master/evaluation/lfw.nn4.v1.epoch-177/accuracies.txt).
|
||||
|
||||
| Image 1 | Image 2 | Distance |
|
||||
|---|---|---|
|
||||
| Lennon 1 | Lennon 2 | 0.310 |
|
||||
| Lennon 1 | Clapton 1 | 1.241 |
|
||||
| Lennon 1 | Clapton 2 | 1.056 |
|
||||
| Lennon 2 | Clapton 1 | 1.386 |
|
||||
| Lennon 2 | Clapton 2 | 1.073 |
|
||||
| Clapton 1 | Clapton 2 | 0.259 |
|
||||
| Lennon 1 | Lennon 2 | 0.298 |
|
||||
| Lennon 1 | Clapton 1 | 0.985 |
|
||||
| Lennon 1 | Clapton 2 | 0.855 |
|
||||
| Lennon 2 | Clapton 1 | 0.904 |
|
||||
| Lennon 2 | Clapton 2 | 0.955 |
|
||||
| Clapton 1 | Clapton 2 | 0.712 |
|
||||
|
|
|
|||
|
|
@ -88,7 +88,7 @@ Run the classifier on your images with:
|
|||
|
||||
| Person | Image | Prediction | Confidence |
|
||||
|---|---|---|---|
|
||||
| Carell | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/carell.jpg' width='200px'></img> | SteveCarell | 0.78 |
|
||||
| Adams | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/adams.jpg' width='200px'></img> | AmyAdams | 0.87 |
|
||||
| Lennon 1 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-1.jpg' width='200px'></img> | DavidBoreanaz | 0.28 |
|
||||
| Lennon 2 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-2.jpg' width='200px'></img> | DavidBoreanaz | 0.56 |
|
||||
| Carell | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/carell.jpg' width='200px'></img> | SteveCarell | 0.96 |
|
||||
| Adams | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/adams.jpg' width='200px'></img> | AmyAdams | 0.98 |
|
||||
| Lennon 1 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-1.jpg' width='200px'></img> | DavidBoreanaz | 0.27 |
|
||||
| Lennon 2 (Unknown) | <img src='https://raw.githubusercontent.com/cmusatyalab/openface/master/images/examples/lennon-2.jpg' width='200px'></img> | DavidBoreanaz | 0.43 |
|
||||
|
|
|
|||
|
|
@ -0,0 +1,13 @@
|
|||
fold, threshold, accuracy
|
||||
0, 0.71, 0.86
|
||||
1, 0.71, 0.83
|
||||
2, 0.77, 0.82
|
||||
3, 0.70, 0.84
|
||||
4, 0.68, 0.82
|
||||
5, 0.71, 0.85
|
||||
6, 0.69, 0.85
|
||||
7, 0.71, 0.85
|
||||
8, 0.71, 0.86
|
||||
9, 0.67, 0.88
|
||||
|
||||
avg, 0.8483 +/- 0.0172
|
||||
|
|
@ -1,13 +1,13 @@
|
|||
fold, threshold, accuracy
|
||||
0, 0.71, 0.86
|
||||
1, 0.71, 0.83
|
||||
2, 0.77, 0.82
|
||||
3, 0.70, 0.84
|
||||
4, 0.68, 0.82
|
||||
5, 0.71, 0.85
|
||||
6, 0.69, 0.85
|
||||
7, 0.71, 0.85
|
||||
8, 0.71, 0.86
|
||||
9, 0.67, 0.88
|
||||
0, 0.68, 0.82
|
||||
1, 0.68, 0.81
|
||||
2, 0.68, 0.79
|
||||
3, 0.68, 0.80
|
||||
4, 0.68, 0.81
|
||||
5, 0.68, 0.83
|
||||
6, 0.68, 0.83
|
||||
7, 0.68, 0.82
|
||||
8, 0.68, 0.81
|
||||
9, 0.60, 0.80
|
||||
|
||||
avg, 0.8483 +/- 0.0172
|
||||
avg, 0.8138 +/- 0.0149
|
||||
|
|
|
|||
Binary file not shown.
|
Before Width: | Height: | Size: 57 KiB After Width: | Height: | Size: 59 KiB |
|
|
@ -20,170 +20,92 @@ import os
|
|||
import random
|
||||
import sys
|
||||
|
||||
from skimage import io
|
||||
|
||||
from .. import helper
|
||||
from .. import data
|
||||
|
||||
TEMPLATE = np.float32([
|
||||
(0.0792396913815, 0.339223741112), (0.0829219487236, 0.456955367943),
|
||||
(0.0967927109165, 0.575648016728), (0.122141515615, 0.691921601066),
|
||||
(0.168687863544, 0.800341263616), (0.239789390707, 0.895732504778),
|
||||
(0.325662452515, 0.977068762493), (0.422318282013, 1.04329000149),
|
||||
(0.531777802068, 1.06080371126), (0.641296298053, 1.03981924107),
|
||||
(0.738105872266, 0.972268833998), (0.824444363295, 0.889624082279),
|
||||
(0.894792677532, 0.792494155836), (0.939395486253, 0.681546643421),
|
||||
(0.96111933829, 0.562238253072), (0.970579841181, 0.441758925744),
|
||||
(0.971193274221, 0.322118743967), (0.163846223133, 0.249151738053),
|
||||
(0.21780354657, 0.204255863861), (0.291299351124, 0.192367318323),
|
||||
(0.367460241458, 0.203582210627), (0.4392945113, 0.233135599851),
|
||||
(0.586445962425, 0.228141644834), (0.660152671635, 0.195923841854),
|
||||
(0.737466449096, 0.182360984545), (0.813236546239, 0.192828009114),
|
||||
(0.8707571886, 0.235293377042), (0.51534533827, 0.31863546193),
|
||||
(0.516221448289, 0.396200446263), (0.517118861835, 0.473797687758),
|
||||
(0.51816430343, 0.553157797772), (0.433701156035, 0.604054457668),
|
||||
(0.475501237769, 0.62076344024), (0.520712933176, 0.634268222208),
|
||||
(0.565874114041, 0.618796581487), (0.607054002672, 0.60157671656),
|
||||
(0.252418718401, 0.331052263829), (0.298663015648, 0.302646354002),
|
||||
(0.355749724218, 0.303020650651), (0.403718978315, 0.33867711083),
|
||||
(0.352507175597, 0.349987615384), (0.296791759886, 0.350478978225),
|
||||
(0.631326076346, 0.334136672344), (0.679073381078, 0.29645404267),
|
||||
(0.73597236153, 0.294721285802), (0.782865376271, 0.321305281656),
|
||||
(0.740312274764, 0.341849376713), (0.68499850091, 0.343734332172),
|
||||
(0.353167761422, 0.746189164237), (0.414587777921, 0.719053835073),
|
||||
(0.477677654595, 0.706835892494), (0.522732900812, 0.717092275768),
|
||||
(0.569832064287, 0.705414478982), (0.635195811927, 0.71565572516),
|
||||
(0.69951672331, 0.739419187253), (0.639447159575, 0.805236879972),
|
||||
(0.576410514055, 0.835436670169), (0.525398405766, 0.841706377792),
|
||||
(0.47641545769, 0.837505914975), (0.41379548902, 0.810045601727),
|
||||
(0.380084785646, 0.749979603086), (0.477955996282, 0.74513234612),
|
||||
(0.523389793327, 0.748924302636), (0.571057789237, 0.74332894691),
|
||||
(0.672409137852, 0.744177032192), (0.572539621444, 0.776609286626),
|
||||
(0.5240106503, 0.783370783245), (0.477561227414, 0.778476346951)])
|
||||
|
||||
TPL_MIN, TPL_MAX = np.min(TEMPLATE, axis=0), np.max(TEMPLATE, axis=0)
|
||||
MINMAX_TEMPLATE = (TEMPLATE - TPL_MIN) / (TPL_MAX - TPL_MIN)
|
||||
|
||||
class NaiveDlib:
|
||||
# https://github.com/cmusatyalab/openface/blob/master/images/dlib-landmark-mean.png
|
||||
INNER_EYES_AND_BOTTOM_LIP = np.array([39, 42, 57])
|
||||
OUTER_EYES_AND_NOSE = np.array([36, 45, 33])
|
||||
|
||||
def __init__(self, faceMean, facePredictor):
|
||||
def __init__(self, facePredictor):
|
||||
"""Initialize the dlib-based alignment."""
|
||||
self.detector = dlib.get_frontal_face_detector()
|
||||
self.normMeanAlignPoints = loadMeanPoints(faceMean)
|
||||
self.predictor = dlib.shape_predictor(facePredictor)
|
||||
|
||||
def getAllFaceBoundingBoxes(self, img):
|
||||
return self.detector(img, 1)
|
||||
def getAllFaceBoundingBoxes(self, rgbImg):
|
||||
try:
|
||||
return self.detector(rgbImg, 1)
|
||||
except Exception as e:
|
||||
print("Warning: {}".format(e))
|
||||
# In rare cases, exceptions are thrown.
|
||||
return []
|
||||
|
||||
def getLargestFaceBoundingBox(self, img):
|
||||
faces = self.detector(img, 1)
|
||||
def getLargestFaceBoundingBox(self, rgbImg):
|
||||
faces = self.getAllFaceBoundingBoxes(rgbImg)
|
||||
if len(faces) > 0:
|
||||
return max(faces, key=lambda rect: rect.width() * rect.height())
|
||||
|
||||
def align(self, img, bb):
|
||||
points = self.predictor(img, bb)
|
||||
def align(self, rgbImg, bb):
|
||||
points = self.predictor(rgbImg, bb)
|
||||
return list(map(lambda p: (p.x, p.y), points.parts()))
|
||||
|
||||
def alignImg(self, method, size, img, bb=None,
|
||||
outputPrefix=None, outputDebug=False,
|
||||
expandBox=False, alignPoints=None):
|
||||
if outputPrefix:
|
||||
helper.mkdirP(os.path.dirname(outputPrefix))
|
||||
|
||||
def getName(tag=None):
|
||||
if tag is None:
|
||||
return "{}.png".format(outputPrefix)
|
||||
else:
|
||||
return "{}-{}.png".format(outputPrefix, tag)
|
||||
|
||||
def alignImg(self, method, size, rgbImg, bb=None,
|
||||
landmarks=None, landmarkIndices=INNER_EYES_AND_BOTTOM_LIP):
|
||||
if bb is None:
|
||||
try:
|
||||
bb = self.getLargestFaceBoundingBox(img)
|
||||
except Exception as e:
|
||||
print("Warning: {}".format(e))
|
||||
# In rare cases, exceptions are thrown.
|
||||
return
|
||||
bb = self.getLargestFaceBoundingBox(rgbImg)
|
||||
if bb is None:
|
||||
# Most failed detection attempts return here.
|
||||
return
|
||||
|
||||
if alignPoints is None:
|
||||
alignPoints = self.align(img, bb)
|
||||
meanAlignPoints = transformPoints(self.normMeanAlignPoints, bb, True)
|
||||
if landmarks is None:
|
||||
landmarks = self.align(rgbImg, bb)
|
||||
|
||||
(xs, ys) = zip(*meanAlignPoints)
|
||||
tightBb = dlib.rectangle(left=min(xs), right=max(xs),
|
||||
top=min(ys), bottom=max(ys))
|
||||
npLandmarks = np.float32(landmarks)
|
||||
|
||||
if method != 'tightcrop':
|
||||
npAlignPoints = np.float32(alignPoints)
|
||||
npMeanAlignPoints = np.float32(meanAlignPoints)
|
||||
|
||||
if method == 'tightcrop':
|
||||
warpedImg = img
|
||||
elif method == 'affine':
|
||||
ss = np.array([39, 42, 57]) # Eyes and bottom lip.
|
||||
npAlignPointsSS = npAlignPoints[ss]
|
||||
npMeanAlignPointsSS = npMeanAlignPoints[ss]
|
||||
H = cv2.getAffineTransform(npAlignPointsSS, npMeanAlignPointsSS)
|
||||
warpedImg = cv2.warpAffine(img, H, np.shape(img)[0:2])
|
||||
elif method == 'perspective':
|
||||
ss = np.array([39, 42, 48, 54]) # Eyes and corners of mouth.
|
||||
npAlignPointsSS = npAlignPoints[ss]
|
||||
npMeanAlignPointsSS = npMeanAlignPoints[ss]
|
||||
H = cv2.getPerspectiveTransform(
|
||||
npAlignPointsSS, npMeanAlignPointsSS)
|
||||
warpedImg = cv2.warpPerspective(img, H, np.shape(img)[0:2])
|
||||
elif method == 'homography':
|
||||
(H, mask) = cv2.findHomography(npAlignPoints, npMeanAlignPoints,
|
||||
method=cv2.LMEDS)
|
||||
warpedImg = cv2.warpPerspective(img, H, np.shape(img)[0:2])
|
||||
if method == 'affine':
|
||||
H = cv2.getAffineTransform(npLandmarks[landmarkIndices],
|
||||
size*MINMAX_TEMPLATE[landmarkIndices])
|
||||
thumbnail = cv2.warpAffine(rgbImg, H, (size, size))
|
||||
else:
|
||||
print("Error: method '{}' is unimplemented.".format(method))
|
||||
sys.exit(-1)
|
||||
raise Exception('Unrecognized method: {}'.format(method))
|
||||
|
||||
if method == 'tightcrop':
|
||||
wAlignPoints = alignPoints
|
||||
else:
|
||||
wBb = self.getLargestFaceBoundingBox(warpedImg)
|
||||
if wBb is None:
|
||||
return
|
||||
wAlignPoints = self.align(warpedImg, wBb)
|
||||
wMeanAlignPoints = transformPoints(
|
||||
self.normMeanAlignPoints, wBb, True)
|
||||
|
||||
if outputDebug:
|
||||
annotatedImg = annotate(img, bb, alignPoints, meanAlignPoints)
|
||||
io.imsave(getName("orig"), img)
|
||||
io.imsave(getName("annotated"), annotatedImg)
|
||||
|
||||
if method != 'tightcrop':
|
||||
wAnnotatedImg = annotate(warpedImg, wBb,
|
||||
wAlignPoints, wMeanAlignPoints)
|
||||
io.imsave(getName("warped"), warpedImg)
|
||||
io.imsave(getName("warped-annotated"), wAnnotatedImg)
|
||||
|
||||
if len(warpedImg.shape) != 3:
|
||||
print(" + Warning: Result does not have 3 dimensions.")
|
||||
return None
|
||||
|
||||
(xs, ys) = zip(*wAlignPoints)
|
||||
xRange = max(xs) - min(xs)
|
||||
yRange = max(ys) - min(ys)
|
||||
if expandBox:
|
||||
(l, r, t, b) = (min(xs) - 0.20 * xRange, max(xs) + 0.20 * xRange,
|
||||
min(ys) - 0.65 * yRange, max(ys) + 0.20 * yRange)
|
||||
else:
|
||||
(l, r, t, b) = (min(xs), max(xs), min(ys), max(ys))
|
||||
(w, h, _) = warpedImg.shape
|
||||
if 0 <= l <= w and 0 <= r <= w and 0 <= b <= h and 0 <= t <= h:
|
||||
cwImg = cv2.resize(warpedImg[t:b, l:r], (size, size))
|
||||
h, edges = np.histogram(cwImg.ravel(), 16, [0, 256])
|
||||
s = sum(h)
|
||||
if any(h > 0.65 * s):
|
||||
print("Warning: Image is likely a single color.")
|
||||
return
|
||||
else:
|
||||
print("Warning: Unable to align and crop to the "
|
||||
"face's bounding box.")
|
||||
return
|
||||
|
||||
if outputDebug:
|
||||
io.imsave(getName(), cwImg)
|
||||
return cwImg
|
||||
|
||||
|
||||
def transformPoints(points, bb, toImgCoords):
|
||||
if toImgCoords:
|
||||
def scale(p):
|
||||
(x, y) = p
|
||||
return (int((x * bb.width()) + bb.left()),
|
||||
int((y * bb.height()) + bb.top()))
|
||||
else:
|
||||
def scale(p):
|
||||
(x, y) = p
|
||||
return (float(x - bb.left()) / bb.width(),
|
||||
float(y - bb.top()) / bb.height())
|
||||
return list(map(scale, points))
|
||||
|
||||
|
||||
def loadMeanPoints(modelFname):
|
||||
def parse(line):
|
||||
(x, y) = line.strip().split(",")
|
||||
return (float(x), float(y))
|
||||
with open(modelFname, 'r') as f:
|
||||
return [parse(line) for line in f]
|
||||
|
||||
|
||||
def annotate(img, box, points=None, meanPoints=None):
|
||||
a = np.copy(img)
|
||||
bl = (box.left(), box.bottom())
|
||||
tr = (box.right(), box.top())
|
||||
cv2.rectangle(a, bl, tr, color=(153, 255, 204), thickness=3)
|
||||
for p in points:
|
||||
cv2.circle(a, center=p, radius=3, color=(102, 204, 255), thickness=-1)
|
||||
for p in meanPoints:
|
||||
cv2.circle(a, center=p, radius=3, color=(0, 0, 0), thickness=-1)
|
||||
return a
|
||||
return thumbnail
|
||||
|
|
|
|||
|
|
@ -14,8 +14,7 @@
|
|||
|
||||
import os
|
||||
|
||||
from skimage import io
|
||||
|
||||
import cv2
|
||||
|
||||
class Image:
|
||||
|
||||
|
|
@ -25,17 +24,20 @@ class Image:
|
|||
self.path = path
|
||||
self.rgb = None
|
||||
|
||||
def getRGB(self, cache=False):
|
||||
if self.rgb is not None:
|
||||
return self.rgb
|
||||
def getBGR(self):
|
||||
try:
|
||||
bgr = cv2.imread(self.path)
|
||||
except:
|
||||
bgr = None
|
||||
return bgr
|
||||
|
||||
def getRGB(self):
|
||||
bgr = self.getBGR()
|
||||
if bgr is not None:
|
||||
rgb = cv2.cvtColor(bgr, cv2.COLOR_BGR2RGB)
|
||||
else:
|
||||
try:
|
||||
rgb = io.imread(self.path)
|
||||
except:
|
||||
rgb = None
|
||||
if cache:
|
||||
self.rgb = rgb
|
||||
return rgb
|
||||
rgb = None
|
||||
return rgb
|
||||
|
||||
def __repr__(self):
|
||||
return "({}, {})".format(self.cls, self.name)
|
||||
|
|
|
|||
|
|
@ -24,8 +24,6 @@ import cv2
|
|||
import random
|
||||
import shutil
|
||||
|
||||
from skimage import io
|
||||
|
||||
modelDir = os.path.join(fileDir, '..', 'models')
|
||||
dlibModelDir = os.path.join(modelDir, 'dlib')
|
||||
openfaceModelDir = os.path.join(modelDir, 'openface')
|
||||
|
|
@ -84,7 +82,14 @@ def alignMain(args):
|
|||
# Shuffle so multiple versions can be run at once.
|
||||
random.shuffle(imgs)
|
||||
|
||||
align = NaiveDlib(args.dlibFaceMean, args.dlibFacePredictor)
|
||||
if args.landmarks == 'outerEyesAndNose':
|
||||
landmarkIndices = NaiveDlib.OUTER_EYES_AND_NOSE
|
||||
elif args.landmarks == 'innerEyesAndBottomLip':
|
||||
landmarkIndices = NaiveDlib.INNER_EYES_AND_BOTTOM_LIP
|
||||
else:
|
||||
raise Exception("Landmarks unrecognized: {}".format(args.landmarks))
|
||||
|
||||
align = NaiveDlib(args.dlibFacePredictor)
|
||||
|
||||
nFallbacks = 0
|
||||
for imgObject in imgs:
|
||||
|
|
@ -94,11 +99,10 @@ def alignMain(args):
|
|||
imgName = outputPrefix + ".png"
|
||||
|
||||
if not os.path.isfile(imgName):
|
||||
rgb = imgObject.getRGB(cache=False)
|
||||
out = align.alignImg(args.method, args.size, rgb,
|
||||
outputPrefix=outputPrefix,
|
||||
outputDebug=args.outputDebugImages)
|
||||
if args.fallbackLfw and out is None:
|
||||
rgb = imgObject.getRGB()
|
||||
outRgb = align.alignImg('affine', args.size, rgb,
|
||||
landmarkIndices = landmarkIndices)
|
||||
if args.fallbackLfw and outRgb is None:
|
||||
nFallbacks += 1
|
||||
deepFunneled = "{}/{}.jpg".format(os.path.join(args.fallbackLfw,
|
||||
imgObject.cls),
|
||||
|
|
@ -107,9 +111,12 @@ def alignMain(args):
|
|||
imgObject.cls),
|
||||
imgObject.name))
|
||||
|
||||
if out is not None:
|
||||
io.imsave(imgName, out)
|
||||
print('nFallbacks:', nFallbacks)
|
||||
if outRgb is not None:
|
||||
outBgr = cv2.cvtColor(outRgb, cv2.COLOR_RGB2BGR)
|
||||
cv2.imwrite(imgName, outBgr)
|
||||
|
||||
if args.fallbackLfw:
|
||||
print('nFallbacks:', nFallbacks)
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser()
|
||||
|
|
@ -131,22 +138,19 @@ if __name__ == '__main__':
|
|||
default=0) # <= 0 ===> all imgs
|
||||
alignmentParser = subparsers.add_parser(
|
||||
'align', help='Align a directory of images.')
|
||||
alignmentParser.add_argument('method', type=str,
|
||||
choices=['tightcrop', 'affine',
|
||||
'perspective', 'homography'],
|
||||
help="Alignment method.")
|
||||
alignmentParser.add_argument('landmarks', type=str,
|
||||
choices=['outerEyesAndNose', 'innerEyesAndBottomLip'],
|
||||
help='The landmarks to align to.')
|
||||
alignmentParser.add_argument(
|
||||
'outputDir', type=str, help="Output directory of aligned images.")
|
||||
alignmentParser.add_argument('--outputDebugImages', action='store_true',
|
||||
help='Output annotated images for debugging and presenting.')
|
||||
alignmentParser.add_argument('--size', type=int, help="Default image size.",
|
||||
default=152)
|
||||
default=96)
|
||||
alignmentParser.add_argument('--fallbackLfw', type=str,
|
||||
help="If alignment doesn't work, fallback to copying the deep funneled version from this directory..")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
sys.path.append(args.dlibRoot)
|
||||
sys.path = [args.dlibRoot] + sys.path
|
||||
import openface
|
||||
import openface.helper
|
||||
from openface.data import iterImgs
|
||||
|
|
|
|||
|
|
@ -0,0 +1,79 @@
|
|||
#!/usr/bin/env python2
|
||||
#
|
||||
# Copyright 2015 Carnegie Mellon University
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
import os
|
||||
import sys
|
||||
fileDir = os.path.dirname(os.path.realpath(__file__))
|
||||
sys.path.append(os.path.join(fileDir, ".."))
|
||||
|
||||
import argparse
|
||||
import cv2
|
||||
|
||||
modelDir = os.path.join(fileDir, '..', 'models')
|
||||
dlibModelDir = os.path.join(modelDir, 'dlib')
|
||||
openfaceModelDir = os.path.join(modelDir, 'openface')
|
||||
|
||||
|
||||
def main(args):
|
||||
align = NaiveDlib(args.dlibFacePredictor)
|
||||
|
||||
bgrImg = cv2.imread(args.img)
|
||||
if bgrImg is None:
|
||||
raise Exception("Unable to load image: {}".format(imgPath))
|
||||
rgbImg = cv2.cvtColor(bgrImg, cv2.COLOR_BGR2RGB)
|
||||
|
||||
bb = align.getLargestFaceBoundingBox(rgbImg)
|
||||
if bb is None:
|
||||
raise Exception("Unable to find a face: {}".format(imgPath))
|
||||
|
||||
landmarks = align.align(rgbImg, bb)
|
||||
if landmarks is None:
|
||||
raise Exception("Unable to align image: {}".format(imgPath))
|
||||
alignedFace = align.alignImg("affine", args.size, rgbImg, bb, landmarks)
|
||||
|
||||
bl = (bb.left(), bb.bottom())
|
||||
tr = (bb.right(), bb.top())
|
||||
cv2.rectangle(bgrImg, bl, tr, color=(153, 255, 204), thickness=3)
|
||||
for landmark in landmarks:
|
||||
cv2.circle(bgrImg, center=landmark, radius=3, color=(102, 204, 255), thickness=-1)
|
||||
print("Saving image to 'annotated.png'")
|
||||
cv2.imwrite("annotated.png", bgrImg)
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = argparse.ArgumentParser()
|
||||
|
||||
parser.add_argument('img', type=str, help="Input image.")
|
||||
parser.add_argument('--dlibFacePredictor', type=str, help="Path to dlib's face predictor.",
|
||||
default=os.path.join(dlibModelDir, "shape_predictor_68_face_landmarks.dat"))
|
||||
parser.add_argument('--dlibRoot', type=str,
|
||||
default=os.path.expanduser(
|
||||
"~/src/dlib-18.16/python_examples"),
|
||||
help="dlib directory with the dlib.so Python library.")
|
||||
|
||||
parser.add_argument('landmarks', type=str,
|
||||
choices=['outerEyesAndNose', 'innerEyesAndBottomLip'],
|
||||
help='The landmarks to align to.')
|
||||
parser.add_argument('--size', type=int, help="Default image size.",
|
||||
default=96)
|
||||
args = parser.parse_args()
|
||||
|
||||
sys.path = [args.dlibRoot] + sys.path
|
||||
import openface
|
||||
import openface.helper
|
||||
from openface.data import iterImgs
|
||||
from openface.alignment import NaiveDlib
|
||||
|
||||
main(args)
|
||||
Loading…
Reference in New Issue