App on updating

app.py

@@ -1,30 +1,120 @@
 # -*- coding: utf-8 -*-
-"""app creation.ipynb
+"""xai_app.ipynb
 
 Automatically generated by Colaboratory.
 
 Original file is located at
-    https://colab.research.google.com/drive/1c8HIdMTAJxhNiPY7_kmzP78yFiwdhh8Q
+    https://colab.research.google.com/drive/1XRB_m0JRoi0KugiHw5WSPJzV0udfU69O
 """
 
+# !nvidia-smi
+
+# Commented out IPython magic to ensure Python compatibility.
+# %reload_ext autoreload
+# %autoreload 2
+# %matplotlib inline
+
+
+# Commented out IPython magic to ensure Python compatibility.
+# %cd /content/drive/MyDrive/Bengali Fish Recognizer/
+
+
+import cv2
+import torch
+import numpy as np
 import gradio as gr
-from fastai import *
+import matplotlib as plt
 from fastai.vision.all import *
+from torchvision import transforms
+from torchvision.io import read_image
+from pytorch_grad_cam import GradCAM
+from pytorch_grad_cam.utils.image import show_cam_on_image
+from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
+
+import pathlib
+temp = pathlib.PosixPath
+pathlib.PosixPath = pathlib.WindowsPath
 
-# import pathlib
-# temp = pathlib.PosixPath
-# pathlib.PosixPath = pathlib.WindowsPath
+model = load_learner("models/recognizer_model.pkl")
 
-model = load_learner("models/recgonizer_model.pkl")
+# Transforming to pytorch model
+pytorch_model = model.eval()
 
 labels = ['Ayre', 'Catla', 'Chital', 'Ilish', 'Kachki', 'Kajoli', 'Koi', 'Magur', 'Mola Dhela', 'Mrigal', 'Pabda', 'Pangash', 'Poa', 'Puti', 'Rui', 'Shing', 'Silver Carp', 'Taki', 'Telapia', 'Tengra']
 
-def recognize_image(image_path):
+def xai_visualization(image, image_tensor, targeted_category, model, target_layers):
+
+  cam = GradCAM(model = model, target_layers = target_layers)
+
+  targets = [ClassifierOutputTarget(targeted_category)]
+
+  grayscale_cam = cam(input_tensor = image_tensor, targets = targets)
+
+  mask = grayscale_cam[0, :]
+
+  plt.imshow(image)
+
+  plt.imshow(mask*255, cmap="plasma", alpha=0.6)
+
+  plt.show()
+
+def preprocess_image(image_path):
+  # Reading image
+  # image = cv2.imread(image_path)
+  image = torchvision.io.read_image(image_path)
+
+  # Resizing an image
+  image = cv2.resize(
+      image,
+      dsize=(224, 224),
+      interpolation=cv2.INTER_CUBIC)
+
+  # Converting image to tensor
+  img_tensor = transforms.ToTensor()(image)
+
+  # Converting image to batch
+  img_tensor = img_tensor.reshape(1,3,224,224)
+
+  return image, img_tensor
+
+def target_layers_finding(model):
+  # Available layers
+  layers = list(model.named_modules())
+
+  # For Resnet-50
+  target_layers = [layers[len(layers)-20][1]]
+
+  return target_layers
+
+def classify_image(image_path):
+
+  # Model Prediction
   label, _, probs = model.predict(image_path)
-  return dict(zip(labels, map(float, probs)))
 
-inputs = gr.inputs.Image(shape=(224,224))
-outputs = gr.outputs.Label()
+  # Predicted Category
+  targeted_category = np.argmax(probs)
+
+  # Preprocessed image and image tensor
+  image, img_tensor = preprocess_image(image_path)
+
+  # Target layer
+  target_layer = target_layers_finding(pytorch_model)
+
+  xai_visualization(image, img_tensor, targeted_category, pytorch_model, target_layer)
+
+  # print(f"Category with most probability: {np.argmax(probs)}")
+
+  return image_path, dict(zip(labels, map(float, probs)))
+
+# classify_image('test images/unknown_01.jpg')
+
+
+inputs = gr.Image()
+
+outputs = [
+    gr.Image(),
+    gr.Label(num_top_classes=5)
+  ]
 
 examples = [
     'test images/unknown_01.jpg',
@@ -46,15 +136,11 @@ examples = [
     'test images/unknown_17.jpg'
 ]
 
-interface = gr.Interface(fn=recognize_image, inputs = inputs, outputs=outputs, examples = examples)
+interface = gr.Interface(
+    fn = classify_image,
+    inputs = inputs,
+    outputs = outputs,
+    examples = examples
+  )
 
 interface.launch()
-
-
-
-
-
-
-
-
-

previous_files/app.py

@@ -0,0 +1,126 @@
+# # -*- coding: utf-8 -*-
+# """app creation.ipynb
+
+# Automatically generated by Colaboratory.
+
+# Original file is located at
+#     https://colab.research.google.com/drive/1c8HIdMTAJxhNiPY7_kmzP78yFiwdhh8Q
+# """
+
+# import gradio as gr
+# from fastai import *
+# from fastai.vision.all import *
+
+# # import pathlib
+# # temp = pathlib.PosixPath
+# # pathlib.PosixPath = pathlib.WindowsPath
+
+# model = load_learner("models/recgonizer_model.pkl")
+
+# labels = ['Ayre', 'Catla', 'Chital', 'Ilish', 'Kachki', 'Kajoli', 'Koi', 'Magur', 'Mola Dhela', 'Mrigal', 'Pabda', 'Pangash', 'Poa', 'Puti', 'Rui', 'Shing', 'Silver Carp', 'Taki', 'Telapia', 'Tengra']
+
+# def recognize_image(image_path):
+#   label, _, probs = model.predict(image_path)
+#   return dict(zip(labels, map(float, probs)))
+
+# inputs = gr.inputs.Image(shape=(224,224))
+# outputs = gr.outputs.Label()
+
+# examples = [
+#     'test images/unknown_01.jpg',
+#     'test images/unknown_02.png',
+#     'test images/unknown_03.jpg',
+#     'test images/unknown_04.jpg',
+#     'test images/unknown_05.jpg',
+#     'test images/unknown_06.jpg',
+#     'test images/unknown_07.jpg',
+#     'test images/unknown_08.jpg',
+#     'test images/unknown_09.jpg',
+#     'test images/unknown_10.jpg',
+#     'test images/unknown_11.jpg',
+#     'test images/unknown_12.png',
+#     'test images/unknown_13.jpg',
+#     'test images/unknown_14.png',
+#     'test images/unknown_15.png',
+#     'test images/unknown_16.png',
+#     'test images/unknown_17.jpg'
+# ]
+
+# interface = gr.Interface(fn=recognize_image, inputs = inputs, outputs=outputs, examples = examples)
+
+# interface.launch()
+
+
+
+
+# -*- coding: utf-8 -*-
+"""app creation.ipynb
+
+Automatically generated by Colaboratory.
+
+Original file is located at
+    https://colab.research.google.com/drive/1c8HIdMTAJxhNiPY7_kmzP78yFiwdhh8Q
+"""
+
+import numpy as np
+import gradio as gr
+from fastai import *
+from fastai.vision.all import *
+
+import pathlib
+temp = pathlib.PosixPath
+pathlib.PosixPath = pathlib.WindowsPath
+
+model = load_learner("models/recgonizer_model.pkl")
+
+labels = ['Ayre', 'Catla', 'Chital', 'Ilish', 'Kachki', 'Kajoli', 'Koi', 'Magur', 'Mola Dhela', 'Mrigal', 'Pabda', 'Pangash', 'Poa', 'Puti', 'Rui', 'Shing', 'Silver Carp', 'Taki', 'Telapia', 'Tengra']
+
+def recognize_image(image_path):
+  label, _, probs = model.predict(image_path)
+  # return dict(zip(labels, map(float, probs)))
+  print(f"Category with most probability: {np.argmax(probs)}")
+  return image_path, dict(zip(labels, map(float, probs)))
+
+# inputs = gr.inputs.Image(shape=(224,224))
+# outputs = gr.outputs.Label()
+
+examples = [
+    'test images/unknown_01.jpg',
+    'test images/unknown_02.png',
+    'test images/unknown_03.jpg',
+    'test images/unknown_04.jpg',
+    'test images/unknown_05.jpg',
+    'test images/unknown_06.jpg',
+    'test images/unknown_07.jpg',
+    'test images/unknown_08.jpg',
+    'test images/unknown_09.jpg',
+    'test images/unknown_10.jpg',
+    'test images/unknown_11.jpg',
+    'test images/unknown_12.png',
+    'test images/unknown_13.jpg',
+    'test images/unknown_14.png',
+    'test images/unknown_15.png',
+    'test images/unknown_16.png',
+    'test images/unknown_17.jpg'
+]
+
+interface = gr.Interface(fn=recognize_image, inputs = gr.Image(), outputs = [gr.Image(height=224, width=224), gr.Label(num_top_classes=5)] , examples = examples)
+
+interface.launch()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+