add scut-fbp beauty predictor

1 files changed, 65 insertions, 4 deletions

diff --git a/megapixels/app/processors/face_beauty.py b/megapixels/app/processors/face_beauty.py
index a1ddd9f8..2e8221b7 100644
--- a/megapixels/app/processors/face_beauty.py
+++ b/megapixels/app/processors/face_beauty.py
@@ -1,3 +1,4 @@
+import sys
 import os
 from os.path import join
 from pathlib import Path
@@ -6,6 +7,20 @@ import math
 import cv2 as cv
 import numpy as np
 import imutils
+import pickle
+
+os.environ['CUDA_VISIBLE_DEVICES'] = ''
+import keras
+from keras.layers import Conv2D, Input, MaxPool2D,Flatten, Dense, Permute, GlobalAveragePooling2D
+from keras.models import Model
+from keras.optimizers import adam
+import os.path
+from keras.models import Sequential
+from keras.applications.resnet50 import ResNet50
+#from keras.applications.resnet50 import Dense
+from keras.layers import Dense
+from keras.optimizers import Adam
+from keras.layers import Dropout
 
 from app.utils import im_utils, logger_utils
 from app.models.bbox import BBox
@@ -18,10 +33,56 @@ class FaceBeauty:
 
   # Estimates beauty using CNN
 
-  def __init__(self):
+  def __init__(self, gpu=-1):
+    # don't really need GPU, CPU is quick enough
     self.log = logger_utils.Logger.getLogger()
-    pass
+    resnet = ResNet50(include_top=False, pooling='avg')
+    self.model = Sequential()
+    self.model.add(resnet)
+    self.model.add(Dense(5, activation='softmax'))
+    self.model.layers[0].trainable = False
+    fp_model = join(cfg.DIR_MODELS_KERAS, 'model-ldl-resnet.h5')
+    self.model.load_weights(fp_model)
+
+
+  def beauty(self, im, bbox_dim):
+    '''Predicts facial "beauty" score based on SCUT-FBP attractiveness labels
+    :param im: (numpy.ndarray) BGR image
+    :param bbox_dim: (BBox) dimensioned BBox
+    :returns (float) 0.0-1.0 with 1 being most attractive
+    '''
+
+    face = bbox_dim.to_xyxy()
+    self.log.debug(f'face: {face}')
+
+    cropped_im = im[face[1]:face[3], face[0]:face[2]]
+
+    im_resized = cv.resize(cropped_im, (224, 224))  # force size
+    im_norm = np.array([(im_resized - 127.5) / 127.5])  # subtract mean
+
+    # forward pass
+    pred = self.model.predict(im_norm)
+
+    # combines score to make final estimate?
+    ldList = pred[0]
+    score = 1 * ldList[0] + 2 * ldList[1] + 3 * ldList[2] + 4 * ldList[3] + 5 * ldList[4]
+    return score/5.0
+
+
+  def score_mapping(self, score):
+    '''(deprecated)
+    '''
 
+    # if score <= 1.9:
+    #   score_mapped = ((4 - 2.5) / (1.9 - 1.0)) * (score-1.0) + 2.5
+    # elif score <= 2.8:
+    #   score_mapped = ((5.5 - 4) / (2.8 - 1.9)) * (score-1.9) + 4
+    # elif score <= 3.4:
+    #   score_mapped = ((6.5 - 5.5) / (3.4 - 2.8)) * (score-2.8) + 5.5
+    # elif score <= 4:
+    #   score_mapped = ((8 - 6.5) / (4 - 3.4)) * (score-3.4) + 6.5
+    # elif score < 5:
+    #   score_mapped = ((9 - 8) / (5 - 4)) * (score-4) + 8
 
-  def beauty(self):
-    return 0.5
\ No newline at end of file
+    # return score_mapped
+    return False