update misclas_helper

misclas_helper.py

@@ -2,6 +2,7 @@ import os
 import math
 import numpy as np
 import pandas as pd
+import seaborn as sn
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -24,10 +25,9 @@ from PIL import Image
 from pytorch_grad_cam import GradCAM
 from pytorch_grad_cam.utils.image import show_cam_on_image
 
-# Denormalize the data using test mean and std deviation
 inv_normalize = transforms.Normalize(
-    mean=[-0.50/0.23, -0.50/0.23, -0.50/0.23],
-    std=[1/0.23, 1/0.23, 1/0.23]
+  mean=[-0.50/0.23, -0.50/0.23, -0.50/0.23],
+  std=[1/0.23, 1/0.23, 1/0.23]
 )
 
 
@@ -138,3 +138,82 @@ def display_cifar_misclassified_data(data: list,
 
 # Plot the misclassified data
 
+
+def crop_image_pil2(image): #Crop image with 1:1 output aspect ratio
+
+    image = Image.fromarray(image)
+    print("image type = ", type(Image)) 
+    width, height = image.size
+    if width == height:
+        return image
+    offset  = int(abs(height-width)/2) 
+    if width>height:
+        image = image.crop([offset,0,width-offset,height])
+    else:
+        image = image.crop([0,offset,width,height-offset]) 
+    return image
+
+def resize_image_pil2(image, new_width, new_height):
+    # Convert to PIL image
+    img = crop_image_pil2(image)
+    img = Image.fromarray(np.array(img))
+    # Get original size
+    width, height = img.size
+
+    # Calculate scale
+    width_scale = new_width / width    # RAJA see if this can be deleted
+    height_scale = new_height / height # RAJA see if this can be deleted
+    # Resize
+    # resized = img.resize((int(width*width_scale), int(height*height_scale)), Image.NEAREST)
+    resized = img.resize((32, 32), Image.NEAREST)
+    # Crop to exact size
+    return resized
+    
+def classify_images(list_images, model, device):
+    """
+    Function to run the model on test set and return misclassified images
+    :param model: Network Architecture
+    :param device: CPU/GPU
+    :param test_loader: DataLoader for test set
+    """
+
+    test_transforms = torchvision.transforms.Compose(
+        [
+            torchvision.transforms.ToTensor(),
+            cifar10_normalization(),
+        ]
+    )
+    # Prepare the model for evaluation i.e. drop the dropout layer
+    model.eval()
+    # List to store misclassified Images
+    classified_data = []
+
+    # Reset the gradients
+    with torch.no_grad():
+        # Extract images, labels in a batch
+        for image in list_images:
+          #print("image type = ", type(image))
+          orig_image = image
+          if(image is None):
+            pred = 10 #This entry indicates none in classes, empty string
+          else:
+            #print("before resize image shape = ", image.shape)
+            image = resize_image_pil2(image, 32, 32)
+            image = np.asarray(image)
+            #print("numpy image dtype = ", image.dtype)
+            #print("before test_transforms image shape = ", image.shape)
+            image = test_transforms(image)
+            #print("after test_transforms image shape = ", image.shape)
+
+            image = image.unsqueeze(0)
+            #print("after squeeze image shape = ", image.shape)
+
+            # Get the model prediction on the image
+            output = model(image)
+
+            # Convert the output from one-hot encoding to a value
+            pred = output.argmax(dim=1, keepdim=True)
+
+          classified_data.append((orig_image, pred))
+
+    return classified_data