import torch
from torchvision import transforms
import numpy as np
import gradio as gr
from PIL import Image
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.image import show_cam_on_image
from custom_resnet import *
#from resnet import ResNet18  # Assuming you have a custom ResNet18 implementation

def load_custom_state_dict(model, state_dict):
    model_state_dict = model.state_dict()
    # Filter out unexpected keys
    filtered_state_dict = {k: v for k, v in state_dict.items() if k in model_state_dict}
    # Update the model's state_dict
    model_state_dict.update(filtered_state_dict)
    # Load the updated state_dict to the model
    model.load_state_dict(model_state_dict)

    
model = CustomResNet()  # Replace this with your CustomResNet if necessary
# Load the state_dict using the custom function
state_dict = torch.load("model_pth.ckpt", map_location=torch.device('cpu'))
load_custom_state_dict(model, state_dict['state_dict'])

inv_normalize = transforms.Normalize(
    mean=[-0.494 / 0.2470, -0.4822 / 0.2435, -0.4465 / 0.2616],
    std=[1 / 0.2470, 1 / 0.2435, 1 / 0.2616]
)
classes = ('plane', 'car', 'bird', 'cat', 'deer',
           'dog', 'frog', 'horse', 'ship', 'truck')

def inference(input_img, transparency=0.5, target_layer_number=-1, num_images=1, num_top_classes=3):
    transform = transforms.ToTensor()
    org_img = input_img
    input_img = transform(input_img)
    input_img = input_img.unsqueeze(0)
    outputs = model(input_img)
    softmax = torch.nn.Softmax(dim=1)
    probabilities = softmax(outputs)
    confidences = {classes[i]: float(probabilities[0, i]) for i in range(10)}

    _, prediction = torch.max(outputs, 1)

    # Get GradCAM for the specified target_layer_number
    target_layers = [model.layer_2[target_layer_number]]
    cam = GradCAM(model=model, target_layers=target_layers, use_cuda=False)
    grayscale_cam = cam(input_tensor=input_img, targets=None)
    grayscale_cam = grayscale_cam[0, :]
    img = input_img.squeeze(0)
    img = inv_normalize(img)
    rgb_img = np.transpose(img, (1, 2, 0))
    rgb_img = rgb_img.numpy()

    # Convert org_img (PIL image) to a NumPy array before performing arithmetic operations
    visualization = show_cam_on_image(org_img / 255, grayscale_cam, use_rgb=True, image_weight=transparency)

    # Create a list to store multiple visualizations


    # # Generate multiple GradCAM visualizations if num_images > 1
    # for _ in range(num_images - 1):
    #     # Get GradCAM for different target_layer_number if provided by the user
    #     if target_layer_number >= -1:
    #         target_layers = [model.layer_2[target_layer_number]]
    #         cam = GradCAM(model=model, target_layers=target_layers, use_cuda=False)
    #         grayscale_cam = cam(input_tensor=input_img, targets=None)
    #         grayscale_cam = grayscale_cam[0, :]

    #     visualization = show_cam_on_image(org_img / 255, grayscale_cam, use_rgb=True, image_weight=transparency)
    #     visualizations.append(visualization)

    # Get top classes based on user input (up to a maximum of 10)
    top_classes = {k: v for k, v in sorted(confidences.items(), key=lambda item: item[1], reverse=True)[:min(num_top_classes, 10)]}

    return top_classes, visualization




title = "CIFAR10 trained on ResNet18 Model with GradCAM"
description = "A simple Gradio interface to infer on ResNet model, and get GradCAM results"
examples = [["car_1.jpg",0.5,-1],["car_2.jpg",0.5,-1],["cat_1.jpg",0.5,-1],["cat_2.jpg",0.5,-1],["dog_1.jpg",0.5,-1],["dog_2.jpg",0.5,-1],["frog_1.jpg",0.5,-1],["frog_2.jpg",0.5,-1],["horse_1.jpg",0.5,-1],["horse_2.jpg",0.5,-1]]
demo = gr.Interface(
    inference,
    inputs = [gr.Image(shape=(32, 32), label="Input Image"),
              gr.Slider(0, 1, value=0.5, label="Opacity of GradCAM"),
              gr.Slider(-2, -1, value=-2, step=1, label="Which Layer?"),
              gr.Number(default=1, label="Number of GradCAM Images to Show"),
              gr.Slider(1, 10, value=3, step=1, label="Number of Top Classes to Show")],
    outputs = [gr.Label(num_top_classes=5), gr.Image(shape=(32, 32), label="Output").style(width=128, height=128)],
    title = title,
    description = description,
    examples = examples,
)


demo.launch()