app.py

import os
import base64
import gradio as gr
import openai
from transformers import pipeline

FLOWER_LABELS = ["daisy", "dandelion", "rose", "sunflower", "tulip"]

vit_classifier = pipeline(
    "image-classification",
    model="./flower-vit-classifier"
)

clip_classifier = pipeline(
    task="zero-shot-image-classification",
    model="openai/clip-vit-large-patch14"
)

client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

def normalize_label(text):
    text = text.strip().lower().replace(".", "").replace(",", "")
    for label in FLOWER_LABELS:
        if label in text:
            return label
    return text

def classify_openai(image_path):
    with open(image_path, "rb") as f:
        b64 = base64.b64encode(f.read()).decode("utf-8")

    prompt = f"""
You are a flower image classifier.

Classify the image into exactly one of these labels:
{FLOWER_LABELS}

Return only one label and nothing else.
"""

    response = client.responses.create(
        model="gpt-4.1-mini",
        input=[{
            "role": "user",
            "content": [
                {"type": "input_text", "text": prompt},
                {"type": "input_image", "image_url": f"data:image/jpeg;base64,{b64}"}
            ]
        }]
    )

    return normalize_label(response.output_text)

def classify_flower(image):
    vit_results = vit_classifier(image)
    vit_output = {x["label"]: float(x["score"]) for x in vit_results}

    clip_results = clip_classifier(image, candidate_labels=FLOWER_LABELS)
    clip_output = {x["label"]: float(x["score"]) for x in clip_results}

    openai_label = classify_openai(image)

    return {
        "your_vit_model": vit_output,
        "clip_zero_shot": clip_output,
        "openai_prediction": openai_label
    }

example_images = [
    ["example_images/train/daisy/2488902131_3417698611_n.jpg"],
    ["example_images/train/dandelion/3554992110_81d8c9b0bd_m.jpg"],
    ["example_images/train/rose/3415176946_248afe9f32.jpg"],
    ["example_images/train/sunflower/4932144003_cbffc89bf0.jpg"],
    ["example_images/train/tulip/4579128789_1561575458_n.jpg"]
]

iface = gr.Interface(
    fn=classify_flower,
    inputs=gr.Image(type="filepath"),
    outputs=gr.JSON(),
    title="Flower Classification Comparison",
    description="Compare a fine-tuned ViT model, CLIP zero-shot, and OpenAI vision on flower images.",
    examples=example_images
)

if __name__ == "__main__":
    iface.launch()