app.py

import torch
import time
from PIL import Image
from torchvision import models, transforms
from transformers import AutoImageProcessor, AutoModelForImageClassification
import urllib.request
import gradio as gr

# ============================================================
# LOAD IMAGENET RESNET50
# ============================================================
resnet = models.resnet50(weights="IMAGENET1K_V2")
resnet.eval()

transform = transforms.Compose([
    transforms.Resize((224,224)),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

labels_url = "https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt"
imagenet_labels = urllib.request.urlopen(labels_url).read().decode("utf-8").split("\n")

dog_indices = list(range(151, 269))
cat_indices = [281, 282, 283, 284, 285]


# ============================================================
# DOG BREED MODEL
# ============================================================
dog_model_name = "prithivMLmods/Dog-Breed-120"
dog_processor = AutoImageProcessor.from_pretrained(dog_model_name)
dog_model = AutoModelForImageClassification.from_pretrained(dog_model_name)


# ============================================================
# CAT BREED MODEL
# ============================================================
cat_model_name = "dima806/67_cat_breeds_image_detection"
cat_processor = AutoImageProcessor.from_pretrained(cat_model_name)
cat_model = AutoModelForImageClassification.from_pretrained(cat_model_name)


# ============================================================
# PIPELINE FUNCTIONS
# ============================================================
def detect_animal_type(image):
    start = time.time()
    img = transform(image).unsqueeze(0)

    with torch.no_grad():
        logits = resnet(img)
        probs = torch.softmax(logits, dim=1)[0]

    idx = probs.argmax().item()
    conf = float(probs[idx])
    latency = (time.time() - start)

    if idx in dog_indices:
        return "dog", imagenet_labels[idx], conf, latency
    elif idx in cat_indices:
        return "cat", imagenet_labels[idx], conf, latency
    return "other", imagenet_labels[idx], conf, latency


def predict_dog_breed(image):
    start = time.time()
    inputs = dog_processor(images=image, return_tensors="pt")
    with torch.no_grad():
        out = dog_model(**inputs)

    probs = torch.softmax(out.logits, dim=1)[0]
    idx = probs.argmax().item()
    latency = (time.time() - start)

    return dog_model.config.id2label[idx], float(probs[idx]), latency


def predict_cat_breed(image):
    start = time.time()
    inputs = cat_processor(images=image, return_tensors="pt")
    with torch.no_grad():
        out = cat_model(**inputs)

    probs = torch.softmax(out.logits, dim=1)[0]
    idx = probs.argmax().item()
    latency = (time.time() - start)

    return cat_model.config.id2label[idx], float(probs[idx]), latency


# ============================================================
# MAIN PIPELINE
# ============================================================
def run_pipeline(input_image):
    if input_image is None:
        return "—", "—", "—", "—", "—", "—", ""

    total_start = time.time()
    image = input_image.convert("RGB")
    logs = []

    # STEP 1 — SPECIES
    animal, base_label, base_conf, t1 = detect_animal_type(image)
    logs.append(f"[Species Detection] {animal.upper()} | {t1:.4f} s")

    # STEP 2 — BREED
    if animal == "dog":
        breed, conf, t2 = predict_dog_breed(image)
        logs.append(f"[Dog Breed Model] {breed} ({conf:.4f}) | {t2:.4f} s")

        total_latency = (time.time() - total_start)
        logs.append(f"[Total Pipeline Latency] {total_latency:.4f} s")

        return (
            animal.title(),
            breed,
            f"{conf:.4f}",
            f"{total_latency:.4f} s",
            f"{breed} ({conf:.4f})",
            "—",
            "\n".join(logs)
        )

    elif animal == "cat":
        breed, conf, t2 = predict_cat_breed(image)
        logs.append(f"[Cat Breed Model] {breed} ({conf:.4f}) | {t2:.4f} s")

        total_latency = (time.time() - total_start)
        logs.append(f"[Total Pipeline Latency] {total_latency:.4f} s")

        return (
            animal.title(),
            breed,
            f"{conf:.4f}",
            f"{total_latency:.4f} s",
            "—",
            f"{breed} ({conf:.4f})",
            "\n".join(logs)
        )

    # OTHER — run both
    else:
        d_breed, d_conf, d_t = predict_dog_breed(image)
        c_breed, c_conf, c_t = predict_cat_breed(image)

        logs.append(f"[Fallback] Dog Model → {d_breed} ({d_conf:.4f}) | {d_t:.4f} s")
        logs.append(f"[Fallback] Cat Model → {c_breed} ({c_conf:.4f}) | {c_t:.4f} s")

        primary_breed = d_breed if d_conf > c_conf else c_breed
        primary_conf = max(d_conf, c_conf)

        total_latency = (time.time() - total_start)
        logs.append(f"[Total Pipeline Latency] {total_latency:.4f} s")

        return (
            "Other",
            primary_breed,
            f"{primary_conf:.4f}",
            f"{total_latency:.4f} s",
            f"{d_breed} ({d_conf:.4f})",
            f"{c_breed} ({c_conf:.4f})",
            "\n".join(logs)
        )


# ============================================================
# GRADIO UI
# ============================================================
with gr.Blocks(theme=gr.themes.Soft(), title="PawCare AI - Pet Identification") as demo:

    gr.Markdown("# 🐾 PawCare AI — Pet Type & Breed Classifier")

    with gr.Row():
        with gr.Column(scale=1):
            inp = gr.Image(type="pil", label="Upload Pet Image", height=350)
            btn = gr.Button("Run Analysis", variant="primary")

        with gr.Column(scale=1):
            gr.Markdown("### 🔍 Prediction Summary")

            with gr.Row():
                animal_box = gr.Textbox(label="Animal Type", interactive=False)
                breed_box = gr.Textbox(label="Primary Predicted Breed", interactive=False)

            with gr.Row():
                conf_box = gr.Textbox(label="Primary Confidence", interactive=False)
                latency_box = gr.Textbox(label="Total Latency", interactive=False)

            with gr.Row():
                dog_box = gr.Textbox(label="Dog Model Output", interactive=False)
                cat_box = gr.Textbox(label="Cat Model Output", interactive=False)

            with gr.Accordion("Detailed Logs (Technical)", open=False):
                logs = gr.Textbox(lines=12, interactive=False)

    btn.click(
        run_pipeline,
        inputs=inp,
        outputs=[animal_box, breed_box, conf_box, latency_box, dog_box, cat_box, logs]
    )

demo.launch(share=True)