def classify_one_image(self,
imgf,
classes=[
'afraid', 'angry', 'disgusted', 'happy',
'neutral', 'sad', 'surprised'
]):
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
transf = transforms.Compose([
transforms.Scale(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
# Face detection
args = {}
args['threshold'] = 0.0
args['window'] = False
args['ignore_multi'] = True
args['grow'] = 10
args['resize'] = True
args['row_resize'] = 512
args['col_resize'] = 512
args['min_proportion'] = 0.1
with tempfile.TemporaryDirectory() as tempdir:
args['o'] = tempdir
face_detector.transform(extract_faces.AttributeDict(args), [imgf])
cropped = Image.open(tempdir + '/' + os.path.basename(imgf))
cropped = transf(cropped)
input_var = Variable(cropped.view(1, *cropped.shape))
if torch.cuda.is_available():
input_var = input_var.cuda()
output = self.model.forward(input_var).cpu().data.numpy()
softmax = np.exp(output) / np.sum(np.exp(output))
clss = np.argmax(softmax)
fig = plt.figure()
plt.imshow(Image.open(imgf))
fig.subplots_adjust(bottom=0.2)
plt.figtext(0.1, 0.05, ', '.join(classes))
plt.figtext(
0.1, 0.10,
', '.join(['{:.3}'.format(a) for a in softmax.reshape(-1)]))
plt.title(classes[clss])
plt.show()
def _inner(arg):
(k, v, emotion, outdir, args) = arg
# Output
args['o'] = outdir + '/' + emotion[k]
face_detector.transform(AttributeDict(args), v)