Sebahadin1234/Facial_Expression_Recognition_Deep_Learning

1import streamlit as st2import torch3import torch.nn as nn4import torch.nn.functional as F5 6from PIL import Image7import numpy as np8 9from pytorch_grad_cam.utils.image import show_cam_on_image10 11import random12import matplotlib.pyplot as plt13import torchvision.transforms as transforms14import cv215 16 17 18import models19 20 21 22def Display_prediction(model_choice, label_dict,):23    # Camera input24    img_file = st.camera_input("📸 Take a photo to classify")25    # Load the selected model26    if model_choice == "CNN":27        model = models.load_cnn_model()28    elif model_choice == "VGG16":29        model = models.load_vgg_model()30    else:31        model = models.load_vit_model()32    33 34 35 36    if img_file is not None:37        image = Image.open(img_file)38 39        # 🔍 Tightly crop the center to focus on the face40        cropped_image = models.tight_center_crop(image, crop_ratio=0.7)41 42        # Show cropped image to user43        st.image(cropped_image, caption="🧠 Tightly Center-Cropped Image")44 45 46        predict = False47        if st.button("🧠 Predict Emotion"):48            predict = True49        if predict:50            st.write("🧠 Predicting...")51 52            input_tensor = models.preprocess_image(cropped_image, model_type=model_choice)53 54 55            # Inference56            with torch.no_grad():57                outputs = model(input_tensor)58                _, predicted = torch.max(outputs, 1)59                predicted_label = predicted.item() + 1  60 61            st.success(f"🧠 Predicted Emotion: **{label_dict[predicted_label]}**")62 63 64            if model_choice == "CNN":65                target_layer = model.conv2  # Adjust to your CNN66 67                # Grad-CAM68                orig, gradcam_img, pred_label = models.apply_gradcam_streamlit(69                    model=model,70                    input_tensor=input_tensor,71                    target_layer=target_layer,72                    class_names=label_dict,73                    true_label=None74                )75 76                st.subheader("🧠 Grad-CAM Visualization")77 78 79 80                # Convert both images to PIL81                orig_img_pil = Image.fromarray((orig * 255).astype(np.uint8))82                heatmap_img_pil = Image.fromarray(gradcam_img)83 84                # Side-by-side view85                st.image([orig_img_pil, heatmap_img_pil], caption=["Original", "Grad-CAM"], width=300)86 87 88 89 90    if st.button("🎲 Show Random Prediction From Test Dataset"):91        model.eval()92 93        test_dataset = models.test_dataset_cnn94 95 96 97 98 99        if model_choice != "CNN":100            test_dataset = models.test_dataset_v101 102        # Pick a truly random image from the whole dataset103 104 105        index_to_label = {i: int(cls) for i, cls in enumerate(test_dataset.classes)}  # test_dataset.classes should be strings like ['1', '2', ..., '6']106 107 108        total_samples = len(test_dataset)109        rand_index = random.randint(0, total_samples - 1)110 111        # Load image and label directly112        image, label = test_dataset[rand_index]113        input_tensor = image.unsqueeze(0)  # Add batch dimension114 115        # Run prediction116        model.eval()117        with torch.no_grad():118            output = model(input_tensor)119            _, predicted = torch.max(output, 1)120 121        # Convert class index (0-based) to folder label (1-based)122        true_label = int(test_dataset.classes[label])123        predicted_label = int(test_dataset.classes[predicted.item()])124 125 126        127 128        # Convert image for display129        image_disp = image.permute(1, 2, 0).cpu().numpy()130        image_disp = image_disp * 0.5 + 0.5  # unnormalize131        image_disp = np.clip(image_disp, 0, 1)132 133        # Display image using Matplotlib134        fig, ax = plt.subplots(figsize=(6, 6))135        ax.imshow(image_disp)136        ax.set_title(f"✅ True: {label_dict[true_label]}\n🤖 Predicted: {label_dict[predicted_label]}")137        ax.axis("off")138        st.pyplot(fig)139 140 141 142 143 144        if model_choice == "CNN":145            # ----------------------------146            # 🧠 Apply Grad-CAM on Selected Random Image147            # ----------------------------148 149        150            target_layer = model.conv2151 152            # Prepare the single image tensor for Grad-CAM153            input_tensor = image.unsqueeze(0)154 155            # Grad-CAM156            img_disp, gradcam_overlay, _ = models.apply_gradcam_streamlit(157                model=model,158                input_tensor=input_tensor,159                target_layer=target_layer,160                class_names=label_dict,161                true_label=true_label162            )163 164            st.subheader("🔥 Grad-CAM on Random Test Image")165 166            # Convert both to displayable format167            orig_pil = Image.fromarray((img_disp * 255).astype(np.uint8))168            heatmap_pil = Image.fromarray(gradcam_overlay)169 170            # Side-by-side in Streamlit171            st.image([orig_pil, heatmap_pil], caption=["Original", "Grad-CAM"], width=300)172