import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from torchvision import models
from torchvision import transforms
from PIL import Image
from PIL import UnidentifiedImageError
import cv2
import segmentation_models_pytorch as sm
import gradio as gr
from aib_cdr import calculate_cdr

class Net1(nn.Module):
    def __init__(self):
        super(Net1, self).__init__()
        self.model = models.resnet50(pretrained=True)

        for param in self.model.parameters():
            param.requires_grad = False
        num_ftrs = self.model.fc.in_features
        self.model.fc = nn.Sequential(
            nn.Linear(num_ftrs, 512),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(512, 3),
            nn.Softmax(dim=1)
        )

    def forward(self, x):
        x = self.model(x)
        return x
class Net2(nn.Module):
    def __init__(self):
        super(Net2, self).__init__()
        self.model = models.resnet50(pretrained=True)

        for param in self.model.parameters():
            param.requires_grad = False
        num_ftrs = self.model.fc.in_features
        self.model.fc = nn.Sequential(
            nn.Linear(num_ftrs, 512),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(512, 2)
        )

    def forward(self, x):
        x = self.model(x)
        probas = nn.functional.softmax(x, dim=1)
        return x, probas

disc_model_path = 'models/disc_model.pth'
cup_model_path = 'models/cup_model.pth'
quality_model_path = 'models/quality_model.pth'
camtype_model_path = 'models/camtype_model.pth'
dr_model_path = 'models/dr_model.pth'

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

disc_model = sm.Unet('resnet34', classes=2, activation=None)
disc_model.load_state_dict(torch.load(disc_model_path, map_location=device))
disc_model.to(device)
disc_model.eval()

cup_model = sm.Unet('resnet34', classes=2, activation=None)
cup_model.load_state_dict(torch.load(cup_model_path, map_location=device))
cup_model.to(device)
cup_model.eval()

quality_model = Net1().to(device)
quality_model.load_state_dict(torch.load(quality_model_path,map_location=device))
quality_model.to(device)
quality_model.eval()

camtype_model = models.resnet50(pretrained=True)
num_features = camtype_model.fc.in_features
camtype_model.fc = nn.Linear(num_features, 1)
camtype_model.load_state_dict(torch.load(camtype_model_path,map_location=device))
camtype_model.to(device)
camtype_model.eval()

dr_model = Net2().to(device)
dr_model.load_state_dict(torch.load(dr_model_path,map_location=device))
dr_model.to(device)
dr_model.eval()

transform = transforms.Compose([
    transforms.Resize((512, 512)),
    transforms.ToTensor(),  
])

def model_interface(image):
    image = Image.fromarray(image.astype('uint8'), 'RGB')
    image = image.resize((512, 512))
    image_np = np.array(image)
    image_tensor = transform(image).unsqueeze(0).to(device)

    cdr = calculate_cdr(image_tensor, disc_model, cup_model, device)
    cdr = round(cdr,2)

    quality_probs = quality_model(image_tensor)
    quality_probs = quality_probs.flatten()
    quality_dict = {"Image Quality : ACCEPTABLE": quality_probs[0].item(), "Image Quality : GOOD": quality_probs[1].item(),"Image Quality : POOR" : quality_probs[2].item()}

    camtype_pred = camtype_model(image_tensor)
    camtype_pred = camtype_pred.item()

    camtype = "Eidon" if camtype_pred == 0 else "Nidek"

    dr_grading = dr_model(image_tensor)
    dr_grading = dr_grading[1].flatten()
    dr_dict = {"DR : Negative": dr_grading[0].item(), "DR : Positive": dr_grading[1].item()}

    disc_pred = disc_model(image_tensor)
    disc_mask = (disc_pred.argmax(dim=1) > 0.5).squeeze().cpu().numpy()
    disc_contours, _ = cv2.findContours(disc_mask.astype('uint8'), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    image_np1 = np.array(image)
    image_np1 = cv2.drawContours(image_np1, disc_contours, -1, (0, 255, 0), 2)

    cup_pred = cup_model(image_tensor)
    cup_mask = (cup_pred.argmax(dim=1) > 0.5).squeeze().cpu().numpy()
    cup_contours, _ = cv2.findContours(cup_mask.astype('uint8'), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    image_np2 = np.array(image)
    image_np2 = cv2.drawContours(image_np2, cup_contours, -1, (0, 0, 255), 2)

    return quality_dict, camtype, dr_dict, image_np1 , image_np2,cdr

iface = gr.Interface(
    fn=model_interface,
    inputs=gr.inputs.Image(),
    outputs=[
        gr.outputs.Label(num_top_classes=3, label="Image Quality"),
        gr.outputs.Textbox(label="Camera Type"),
        gr.outputs.Label(num_top_classes=2, label="DR Grading"),
        gr.outputs.Image(type='numpy', label="Disc Mask"),
        gr.outputs.Image(type='numpy', label="Cup Mask"),
        gr.outputs.Textbox(label="Cup-to-disc Ratio")
    ]
)

iface.launch(debug=True)