import torch
import torch.nn as nn
from torchvision import models, transforms
from PIL import Image
import gradio as gr

# ====== 1. Class names ======
classes = [
    "affenpinscher", "afghan_hound", "african", "airedale", "akita", "american_terrier",
    "appenzeller", "australian_cattledog", "australian_terrier", "basenji", "basset_hound",
    "beagle", "bedlington_terrier", "bernese_mountain", "bichon_frise", "blenheim_spaniel",
    "blood_hound", "bluetick", "border_collie", "border_terrier", "borzoi", "boston_bulldog",
    "bouvier", "boxer", "brabancon", "briard", "brittany_spaniel", "bull_mastiff",
    "cairn_terrier", "cardigan_corgi", "caucasian_ovcharka", "cavapoo", "chesapeake_retriever",
    "chihuahua", "chow", "clumber", "cockapoo", "cocker_spaniel", "coonhound", "cotondetulear",
    "curly_retriever", "dachshund", "dalmatian", "dandie_terrier", "dhole", "dingo", "doberman",
    "english_bulldog", "english_hound", "english_mastiff", "english_setter", "english_sheepdog",
    "english_springer", "entlebucher", "eskimo", "flatcoated_retriever", "fox_terrier",
    "french_bulldog", "german_pointer", "germanlonghair_pointer", "germanshepherd",
    "golden_retriever", "gordon_setter", "great_dane", "groenendael", "havanese", "husky",
    "ibizan_hound", "indian_bakharwal", "indian_chippiparai", "indian_gaddi", "indian_greyhound",
    "indian_mastiff", "indian_mudhol", "indian_pariah", "indian_sheepdog", "indian_spitz",
    "irish_setter", "irish_spaniel", "irish_terrier", "irish_wolfhound", "italian_greyhound",
    "japanese_spaniel", "japanese_spitz", "keeshond", "kelpie", "kelpie_australian",
    "kerryblue_terrier", "kombai", "komondor", "kuvasz", "labradoodle", "labrador",
    "lakeland_terrier", "lapphund_finnish", "leonberg", "lhasa", "malamute", "malinois",
    "maltese", "medium_poodle", "mexicanhairless", "miniature_pinscher", "miniature_poodle",
    "mix", "newfoundland", "norfolk_terrier", "norwegian_buhund", "norwegian_elkhound",
    "norwich_terrier", "otterhound", "papillon", "patterdale_terrier", "pekinese", "pembroke",
    "pitbull", "plott_hound", "pomeranian", "pug", "puggle", "pyrenees", "redbone",
    "rottweiler", "russell_terrier", "saluki", "samoyed", "schipperke", "scottish_deerhound",
    "scottish_terrier", "sealyham_terrier", "sharpei", "shepherd_australian", "shetland_sheepdog",
    "shiba", "shihtzu", "silky_terrier", "spanish_waterdog", "staffordshire_bullterrier",
    "standard_poodle", "stbernard", "sussex_spaniel", "swedish_danish", "swiss_mountain",
    "tervuren", "tibetan_mastiff", "tibetan_terrier", "toy_poodle", "toy_terrier", "vizsla",
    "walker_hound", "weimaraner", "welsh_spaniel", "welsh_terrier", "westhighland_terrier",
    "wheaten_terrier", "whippet", "yorkshire_terrier"
]

# ====== 2. Transform (same as training) ======
transform = transforms.Compose([
    transforms.Lambda(lambda image: image.convert('RGB')),
    transforms.Resize((224, 224)),  # fixed size
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5],
                         std=[0.5, 0.5, 0.5])
])

# ====== 3. Load trained model ======
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
in_features = model.fc.in_features
model.fc = nn.Sequential(
    nn.Linear(in_features, 1024),
    nn.ReLU(),
    nn.Dropout(0.4),
    nn.Linear(1024, len(classes))
)
model.load_state_dict(torch.load("best_model.pth", map_location=device))
model.to(device)
model.eval()

# ====== 4. Prediction function ======
def predict_breed(image):
    image = image.convert('RGB')
    img_tensor = transform(image).unsqueeze(0).to(device)
    with torch.no_grad():
        outputs = model(img_tensor)
        probs = torch.softmax(outputs, dim=1)[0]
        top_probs, top_idxs = torch.topk(probs, 3)
    
    results = {classes[idx]: float(prob.item()) for prob, idx in zip(top_probs, top_idxs)}
    return results

# ====== 5. Gradio Interface ======
title = "🐶 Dog Breed Classifier"
description = "Upload a dog image to predict its breed from 157 possible classes."

demo = gr.Interface(
    fn=predict_breed,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=3),
    title=title,
    description=description
)

# ====== 6. Launch ======
demo.launch()