PS | Change age and gender models to VIT

notebooks/facial_age_gender.ipynb

@@ -22,7 +22,7 @@
     {
      "data": {
       "text/plain": [
-       "'/Users/pareshar/Personal/Github/Facial-feature-detector'"
+       "'/Users/pareshar/Personal/Github/temp/Facial-feature-detector'"
       ]
      },
      "execution_count": 2,
@@ -308,6 +308,206 @@
     "df.sort_values(\"file_name\")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Hugging face pre-trained VIT model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/requests/__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.15) or chardet (5.1.0)/charset_normalizer (2.0.12) doesn't match a supported version!\n",
+      "  warnings.warn(\"urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported \"\n",
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/urllib3/connectionpool.py:1045: InsecureRequestWarning: Unverified HTTPS request is being made to host 'huggingface.co'. Adding certificate verification is strongly advised. See: https://urllib3.readthedocs.io/en/1.26.x/advanced-usage.html#ssl-warnings\n",
+      "  warnings.warn(\n",
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/urllib3/connectionpool.py:1045: InsecureRequestWarning: Unverified HTTPS request is being made to host 'huggingface.co'. Adding certificate verification is strongly advised. See: https://urllib3.readthedocs.io/en/1.26.x/advanced-usage.html#ssl-warnings\n",
+      "  warnings.warn(\n",
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/transformers/models/vit/feature_extraction_vit.py:28: FutureWarning: The class ViTFeatureExtractor is deprecated and will be removed in version 5 of Transformers. Please use ViTImageProcessor instead.\n",
+      "  warnings.warn(\n"
+     ]
+    }
+   ],
+   "source": [
+    "# age\n",
+    "\n",
+    "import os\n",
+    "import cv2\n",
+    "from transformers import ViTImageProcessor, ViTForImageClassification\n",
+    "\n",
+    "os.environ[\n",
+    "        \"CURL_CA_BUNDLE\"\n",
+    "    ] = \"\"  # fixes VPN issue when connecting to hugging face hub\n",
+    "\n",
+    "\n",
+    "image = cv2.imread(\"data/4_6_boy.jpg\")\n",
+    "\n",
+    "\n",
+    "# Init model, transforms\n",
+    "model = ViTForImageClassification.from_pretrained('nateraw/vit-age-classifier')\n",
+    "transforms = ViTImageProcessor.from_pretrained('nateraw/vit-age-classifier')\n",
+    "\n",
+    "# Transform our image and pass it through the model\n",
+    "inputs = transforms(image, return_tensors='pt')\n",
+    "output = model(**inputs)\n",
+    "\n",
+    "# Predicted Class probabilities\n",
+    "proba = output.logits.softmax(1)\n",
+    "\n",
+    "# Predicted Classes\n",
+    "preds = proba.argmax(1)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "0.7176125645637512"
+      ]
+     },
+     "execution_count": 24,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "max(proba[0]).item()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'3-9'"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "id2label = {\n",
+    "    0: \"0-2\",\n",
+    "    1: \"3-9\",\n",
+    "    2: \"10-19\",\n",
+    "    3: \"20-29\",\n",
+    "    4: \"30-39\",\n",
+    "    5: \"40-49\",\n",
+    "    6: \"50-59\",\n",
+    "    7: \"60-69\",\n",
+    "    8: \"more than 70\"\n",
+    "  }\n",
+    "\n",
+    "id2label[int(preds)]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 28,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/urllib3/connectionpool.py:1045: InsecureRequestWarning: Unverified HTTPS request is being made to host 'huggingface.co'. Adding certificate verification is strongly advised. See: https://urllib3.readthedocs.io/en/1.26.x/advanced-usage.html#ssl-warnings\n",
+      "  warnings.warn(\n",
+      "/Users/pareshar/.pyenv/versions/3.8.10/lib/python3.8/site-packages/urllib3/connectionpool.py:1045: InsecureRequestWarning: Unverified HTTPS request is being made to host 'huggingface.co'. Adding certificate verification is strongly advised. See: https://urllib3.readthedocs.io/en/1.26.x/advanced-usage.html#ssl-warnings\n",
+      "  warnings.warn(\n"
+     ]
+    }
+   ],
+   "source": [
+    "# gender\n",
+    "\n",
+    "import os\n",
+    "import cv2\n",
+    "from transformers import ViTImageProcessor, ViTForImageClassification\n",
+    "\n",
+    "os.environ[\n",
+    "        \"CURL_CA_BUNDLE\"\n",
+    "    ] = \"\"  # fixes VPN issue when connecting to hugging face hub\n",
+    "\n",
+    "\n",
+    "image = cv2.imread(\"data/gigi_hadid.webp\")\n",
+    "\n",
+    "\n",
+    "# Init model, transforms\n",
+    "model = ViTForImageClassification.from_pretrained('rizvandwiki/gender-classification')\n",
+    "transforms = ViTImageProcessor.from_pretrained('rizvandwiki/gender-classification')\n",
+    "\n",
+    "# Transform our image and pass it through the model\n",
+    "inputs = transforms(image, return_tensors='pt')\n",
+    "output = model(**inputs)\n",
+    "\n",
+    "# Predicted Class probabilities\n",
+    "proba = output.logits.softmax(1)\n",
+    "\n",
+    "# Predicted Classes\n",
+    "preds = proba.argmax(1)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 29,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "0.9677436351776123"
+      ]
+     },
+     "execution_count": 29,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "max(proba[0]).item()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 30,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'female'"
+      ]
+     },
+     "execution_count": 30,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "id2label = {\n",
+    "    0: \"female\",\n",
+    "    1: \"male\",\n",
+    "  }\n",
+    "\n",
+    "id2label[int(preds)]"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -315,6 +515,11 @@
     "# Other\n",
     "- Dataset used to train model: https://talhassner.github.io/home/projects/Adience/Adience-data.html#agegender"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
   }
  ],
  "metadata": {

requirements.txt

@@ -6,3 +6,4 @@ imutils==0.5.4
 pillow==9.4.0
 pyyaml==6.0
 scikit-learn==1.2.2
+transfomers==4.28.1

src/face_demographics.py

@@ -4,6 +4,8 @@ import numpy as np
 import os
 from typing import Tuple
 from src.cv_utils import get_image
+from transformers import ViTImageProcessor, ViTForImageClassification
+import urllib3
 
 
 with open("parameters.yml", "r") as stream:
@@ -18,7 +20,13 @@ class GetFaceDemographics:
         pass
     
     @staticmethod
-    def get_age(blob) -> Tuple:
+    def preprocess_image_for_caffe_cnn(image: np.array):
+        model_mean = (78.4263377603, 87.7689143744, 114.895847746) # taken from the model page on Caffe
+        blob = cv2.dnn.blobFromImage(image, 1.0, (227, 227), model_mean, swapRB=False)
+        return blob
+        
+    @staticmethod
+    def get_age_cnn(blob) -> Tuple:
         age_net = cv2.dnn.readNet(parameters["face_age"]["config"], parameters["face_age"]["model"])
         age_list = ['(0-2)', '(4-6)', '(8-12)', '(15-20)', '(25-32)', '(38-43)', '(48-53)', '(60-100)']
         age_net.setInput(blob)
@@ -29,7 +37,7 @@ class GetFaceDemographics:
         return age, age_confidence_score
     
     @staticmethod
-    def get_gender(blob) -> Tuple:
+    def get_gender_cnn(blob) -> Tuple:
         gender_net = cv2.dnn.readNet(parameters["face_gender"]["config"], parameters["face_gender"]["model"])
         gender_list = ['Male', 'Female']
         gender_net.setInput(blob)
@@ -38,13 +46,54 @@ class GetFaceDemographics:
         gender = gender_list[i]
         gender_confidence_score = gender_preds[0][i]
         return gender, gender_confidence_score
+    
+    @staticmethod
+    def get_age_vit(image: np.array) -> Tuple:
+        os.environ["CURL_CA_BUNDLE"] = ""  # fixes VPN issue when connecting to hugging face hub
+        urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+        id2label = {
+            0: "0-2",
+            1: "3-9",
+            2: "10-19",
+            3: "20-29",
+            4: "30-39",
+            5: "40-49",
+            6: "50-59",
+            7: "60-69",
+            8: "more than 70"
+        }
+        model = ViTForImageClassification.from_pretrained('nateraw/vit-age-classifier')
+        transforms = ViTImageProcessor.from_pretrained('nateraw/vit-age-classifier')
+        inputs = transforms(image, return_tensors='pt')
+        output = model(**inputs)
+        proba = output.logits.softmax(1)
+        preds = proba.argmax(1)  
+        age_confidence_score = max(proba[0]).item()
+        age = id2label[int(preds)]
+        return age, age_confidence_score
+    
+    @staticmethod
+    def get_gender_vit(image: np.array) -> Tuple:
+        os.environ["CURL_CA_BUNDLE"] = ""  # fixes VPN issue when connecting to hugging face hub
+        urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+        id2label = {
+            0: "female",
+            1: "male",
+        }
+        model = ViTForImageClassification.from_pretrained('rizvandwiki/gender-classification')
+        transforms = ViTImageProcessor.from_pretrained('rizvandwiki/gender-classification')
+        inputs = transforms(image, return_tensors='pt')
+        output = model(**inputs)
+        proba = output.logits.softmax(1)
+        preds = proba.argmax(1)  
+        gender_confidence_score = max(proba[0]).item()
+        gender = id2label[int(preds)]
+        return gender, gender_confidence_score
 
     def main(self, image_input) -> dict:
         image = get_image(image_input)      
-        model_mean = (78.4263377603, 87.7689143744, 114.895847746) # taken from the model page on Caffe
-        blob = cv2.dnn.blobFromImage(image, 1.0, (227, 227), model_mean, swapRB=False)
-        age, age_confidence_score = self.get_age(blob)
-        gender, gender_confidence_score = self.get_gender(blob)
+        age, age_confidence_score = self.get_age_vit(image)
+        gender, gender_confidence_score = self.get_gender_vit(image)
         d = {
             "age_range": age,
             "age_confidence": age_confidence_score,
@@ -53,7 +102,6 @@ class GetFaceDemographics:
         }
         return d
 
-
 if __name__ == "__main__":
     path_to_images = "data/"
     image_files = os.listdir(path_to_images)