Add gender age model

README.md

@@ -135,7 +135,7 @@ show_image(img)
 
 ## Remove unwanted objects
 
-Directly remove unwanted objects in images and photos locally.
+Directly remove unwanted objects in images and photos locally. Just click on the object and press the "spacebar" to automatically select and delete the object from the image. Finally, press "s" to save the final image.
 
 ```python
 from abraia.utils import load_image, Sketcher
@@ -161,6 +161,29 @@ sketcher.run(lama.predict)
 
 ![inpaint output](https://github.com/abraia/abraia-multiple/raw/master/images/inpaint-output.jpg)
 
+## Gender Age model
+
+Model to predict gender and age. It can be useful to anonymize minors faces.
+
+```python
+from abraia.faces import Recognition, Attribute
+from abraia.utils import load_image, show_image
+from abraia.draw import render_results
+
+recognition = Recognition()
+attribute = Attribute()
+
+img = load_image('images/image.jpg')
+results = recognition.detect_faces(img)
+faces = recognition.extract_faces(img, results)
+for face, result in zip(faces, results):
+    gender, age, score = attribute.predict(face)
+    result['label'] = f"{gender} {age}"
+    result['confidence'] = score
+img = render_results(img, results)
+show_image(img)
+```
+
 ## Command line interface
 
 The Abraia CLI provides access to the Abraia Cloud Platform through the command line. It makes simple to manage your files and enables bulk image editing capabilities. It provides and easy way to resize, convert, and compress your images - JPEG, WebP, or PNG -, and get them ready to publish on the web. Moreover, you can automatically remove the background, upscale, or anonymize your images in bulk.
@@ -173,7 +196,7 @@ Automatically remove images background and make them transparent in bulk.
 abraia editing removebg "*.jpg"
 ```
 
-![bolt transparent background](https://github.com/abraia/abraia-multiple/raw/master/images/usain-bolt.png)
+![removebg output](https://github.com/abraia/abraia-multiple/raw/master/images/removebg-output.png)
 
 ### Upscale images
 
@@ -203,25 +226,18 @@ Compress images in bulk specifying the input glob pattern or folder:
 abraia editing convert "images/bird*.jpg"
 ```
 
-![Image compressed from url](https://github.com/abraia/abraia-multiple/raw/master/images/birds_o.jpg)
-
-Resize and optimize images maintaining the aspect ratio just specifying the `width` or the `height` of the new image:
-
-```sh
-abraia editing convert images/usain-bolt.jpeg --width 500 
-```
-
-![Usain Bolt resized](https://github.com/abraia/abraia-multiple/raw/master/images/usain-bolt_500.jpeg)
+![image optimized](https://github.com/abraia/abraia-multiple/raw/master/images/birds_optimized.jpg)
 
-You can also automatically change the aspect ratio specifying both `width` and `height` parameters and setting the resize `mode` (pad, crop, thumb):
+Automatically change the aspect ratio specifying both `width` and `height` parameters and setting the resize `mode` (pad, crop, thumb). Or simply resize images maintaining the aspect ratio just specifying the `width` or the `height` of the new image:
 
 ```sh
-abraia editing convert images/lion.jpg --width 333 --height 333 --mode pad 
-abraia editing convert images/lion.jpg --width 333 --height 333
+abraia editing convert images/birds.jpg --width 375 --height 375 --mode pad 
+abraia editing convert images/birds.jpg --width 375 --height 375
+abraia editing convert images/birds.jpg --width 750
 ```
 
-![Image lion smart cropped](https://github.com/abraia/abraia-multiple/raw/master/images/lion_333x333_pad.jpg)
-![Image lion smart cropped](https://github.com/abraia/abraia-multiple/raw/master/images/lion_333x333.jpg)
+![image padded](https://github.com/abraia/abraia-multiple/raw/master/images/birds_padded.jpg)
+![image smart cropped](https://github.com/abraia/abraia-multiple/raw/master/images/birds_cropped.jpg)
 
 So, you can automatically resize all the images in a specific folder preserving the aspect ration of each image just specifying the target `width` or `height`:
 

abraia/detect.py

@@ -4,7 +4,7 @@
 import numpy as np
 import onnxruntime as ort
 
-from .ops import py_cpu_nms, normalize, mask_to_polygon
+from .ops import py_cpu_nms, normalize, mask_to_polygon, softmax
 from .utils import download_file, load_json, get_color, get_providers
 from .utils import load_image, show_image, save_image, Video
 from .draw import render_results
@@ -30,11 +30,6 @@ def preprocess(img):
     return np.expand_dims(img.transpose((2, 0, 1)), axis=0)
 
 
-def softmax(x):
-    e_x = np.exp(x - np.max(x))
-    return e_x / e_x.sum(axis=0)
-
-
 def postprocess(outputs, classes):
     probs = softmax(outputs[0].flatten())
     idx = np.argmax(probs)

abraia/faces/__init__.py

@@ -3,6 +3,7 @@
 from .retinaface import Retinaface
 from .arcface import ArcFace
 from .transform import align_face
+from .genderage import Attribute
 
 
 def euclidean_distance(feat1, feat2):
@@ -35,11 +36,14 @@ def represent_faces(self, img, results=None, size=112):
 
     def identify_faces(self, results, index, threshold=0.45):
         for result in results:
-            sims = [cosine_similarity(result['embeddings'], ind['embeddings']) for ind in index]
-            idx = np.argmax(sims)
-            if sims[idx] > threshold:
-                result['confidence'] = sims[idx]
-                result['label'] = index[idx]['name']
+            del result['confidence']
+            result['label'] = 'unknow'
+            if len(index):
+                sims = [cosine_similarity(result['embeddings'], ind['embeddings']) for ind in index]
+                idx = np.argmax(sims)
+                if sims[idx] > threshold:
+                    result['confidence'] = sims[idx]
+                    result['label'] = index[idx]['name']
         return results
 
 

abraia/faces/genderage.py

@@ -0,0 +1,38 @@
+import cv2
+import numpy as np
+import onnxruntime as ort
+
+from ..utils import download_file
+from ..ops import softmax
+
+
+class Attribute:
+    def __init__(self):
+        """Age and Gender Prediction"""
+        self.input_std = 1.0
+        self.input_mean = 0.0
+
+        model_src = download_file('multiple/models/faces/genderage.simplified.onnx')
+        self.session = ort.InferenceSession(model_src, None)
+
+        inputs = self.session.get_inputs()
+        self.input_size = tuple(inputs[0].shape[2:][::-1])
+
+        self.input_names = [x.name for x in self.session.get_inputs()]
+        self.output_names = [x.name for x in self.session.get_outputs()]
+
+    def preprocess(self, transformed_image):
+        return cv2.dnn.blobFromImage(transformed_image, 1.0 / self.input_std,
+            self.input_size, (self.input_mean, self.input_mean, self.input_mean), swapRB=True)
+
+    def postprocess(self, predictions):
+        scores = softmax(predictions[:2])
+        gender = int(np.argmax(scores))
+        age = int(np.round(predictions[2]*100))
+        return 'Male' if gender == 1 else 'Female', age, float(scores[gender])
+
+    def predict(self, face):
+        blob = self.preprocess(face)
+        predictions = self.session.run(self.output_names, {self.input_names[0]: blob})[0][0]
+        gender, age, score = self.postprocess(predictions)
+        return gender, age, score

abraia/ops.py

@@ -136,3 +136,8 @@ def non_maximum_suppression(objects, iou_threshold):
         idxs = py_cpu_nms(np.array(dets), iou_threshold)
         return [objects[idx] for idx in idxs]
     return []
+
+
+def softmax(x):
+    e_x = np.exp(x - np.max(x))
+    return e_x / e_x.sum(axis=0)

abraia/utils/sketcher.py

@@ -22,7 +22,7 @@ def draw_mask(img, mask, color, opacity = 1):
 
 
 class Sketcher:
-    def __init__(self, img, radius=15):
+    def __init__(self, img, radius=11):
         print(__doc__)
         self.prev_pt = None
         self.win_name = 'Image'