app.py

"""
Aircraft classifier — comparison of three models:
  1. Custom fine-tuned ViT (transfer learning, 6 aircraft classes)
  2. CLIP zero-shot (open-source: openai/clip-vit-large-patch14)
  3. OpenAI GPT-4o vision (closed-source)
"""

import base64
import io
import os

import gradio as gr
from PIL import Image
from transformers import pipeline

CUSTOM_MODEL_ID = "dubattim/aircraft-vit-fs26"
AIRCRAFT_LABELS = [
    "ah64_apache",
    "airbus_a220",
    "airbus_a320",
    "airbus_a321neo",
    "airbus_a330",
    "airbus_a350",
    "airbus_a380",
    "atr_72",
    "b2_spirit",
    "boeing_737",
    "boeing_747",
    "boeing_777",
    "boeing_787_dreamliner",
    "cessna_172",
    "concorde",
    "embraer_e190",
    "f16_fighting_falcon",
    "mig21",
    "sr71_blackbird",
    "v22_osprey",
]
PRETTY = {
    "ah64_apache": "AH-64 Apache",
    "airbus_a220": "Airbus A220",
    "airbus_a320": "Airbus A320",
    "airbus_a321neo": "Airbus A321neo",
    "airbus_a330": "Airbus A330",
    "airbus_a350": "Airbus A350",
    "airbus_a380": "Airbus A380",
    "atr_72": "ATR 72",
    "b2_spirit": "Northrop B-2 Spirit",
    "boeing_737": "Boeing 737",
    "boeing_747": "Boeing 747",
    "boeing_777": "Boeing 777",
    "boeing_787_dreamliner": "Boeing 787 Dreamliner",
    "cessna_172": "Cessna 172",
    "concorde": "Concorde",
    "embraer_e190": "Embraer E190",
    "f16_fighting_falcon": "F-16 Fighting Falcon",
    "mig21": "MiG-21",
    "sr71_blackbird": "SR-71 Blackbird",
    "v22_osprey": "V-22 Osprey",
}

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

try:
    from openai import OpenAI

    openai_client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
except Exception:
    openai_client = None


def _encode_image(path: str) -> str:
    with Image.open(path) as img:
        img = img.convert("RGB")
        buf = io.BytesIO()
        img.save(buf, format="PNG")
        return base64.b64encode(buf.getvalue()).decode("utf-8")


def classify_with_openai(image_path: str) -> dict:
    if openai_client is None or not os.environ.get("OPENAI_API_KEY"):
        return {"error": "OPENAI_API_KEY not configured"}
    b64 = _encode_image(image_path)
    options = ", ".join(PRETTY.values())
    prompt = (
        f"Identify the aircraft in this image. Respond with EXACTLY one of "
        f"these labels and nothing else: {options}."
    )
    resp = openai_client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {
                "role": "user",
                "content": [
                    {"type": "text", "text": prompt},
                    {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{b64}"}},
                ],
            }
        ],
        max_tokens=20,
    )
    answer = resp.choices[0].message.content.strip().lower()
    return {
        PRETTY[k]: (1.0 if PRETTY[k].lower() in answer else 0.0)
        for k in AIRCRAFT_LABELS
    }


def classify(image_path):
    if image_path is None:
        return {}, {}, {}
    custom = {
        PRETTY.get(r["label"], r["label"]): float(r["score"])
        for r in custom_classifier(image_path)
    }
    clip = {
        r["label"]: float(r["score"])
        for r in clip_classifier(image_path, candidate_labels=list(PRETTY.values()))
    }
    openai_out = classify_with_openai(image_path)
    return custom, clip, openai_out


examples = [[f"example_images/{c}.jpg"] for c in AIRCRAFT_LABELS]

with gr.Blocks(title="Aircraft Classifier — Model Comparison") as demo:
    gr.Markdown(
        "# Aircraft Classifier — Model Comparison\n"
        "Upload an aircraft image and compare three approaches: a fine-tuned "
        "ViT (transfer learning), CLIP zero-shot, and OpenAI GPT-4o vision.\n\n"
        f"**Classes:** {', '.join(PRETTY.values())}"
    )
    with gr.Row():
        inp = gr.Image(type="filepath", label="Input image")
        with gr.Column():
            out_custom = gr.Label(label="Custom ViT (fine-tuned)", num_top_classes=5)
            out_clip = gr.Label(label="CLIP zero-shot", num_top_classes=5)
            out_openai = gr.Label(label="OpenAI GPT-4o", num_top_classes=5)
    btn = gr.Button("Classify", variant="primary")
    btn.click(classify, inputs=inp, outputs=[out_custom, out_clip, out_openai])
    gr.Examples(examples=examples, inputs=inp)

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