app/inference.py

# app/inference.py
import os
import io
import json
import base64
import torch
import torch.nn.functional as F
import matplotlib.pyplot as plt
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForImageClassification

# ─────────────────────────────────────────────
# CONFIG
# ─────────────────────────────────────────────
MODEL_REPO = "Arew99/dinov2-costum"  # your Hugging Face repo
ID2NAME_PATH = os.path.join(os.path.dirname(__file__), "id2name.json")

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"🧠 Using device: {DEVICE}")

_model = None
_processor = None
_id2name = None


# ─────────────────────────────────────────────
# HELPER — load id2name mapping
# ─────────────────────────────────────────────
def _load_id2name():
    if os.path.exists(ID2NAME_PATH):
        with open(ID2NAME_PATH, "r") as f:
            data = json.load(f)
        return {int(k): v for k, v in data.items()}
    print("⚠️ id2name.json not found — using placeholder labels.")
    return {i: f"Class {i}" for i in range(101)}


# ─────────────────────────────────────────────
# LOAD MODEL (cached globally)
# ─────────────────────────────────────────────
def load_classification_model():
    global _model, _processor, _id2name

    if _model is not None:
        return _model, _processor, _id2name

    print(f"🔁 Loading model from Hugging Face repo: {MODEL_REPO}")
    _processor = AutoImageProcessor.from_pretrained(MODEL_REPO)
    _model = AutoModelForImageClassification.from_pretrained(
        MODEL_REPO,
        ignore_mismatched_sizes=True,
    ).to(DEVICE)
    _model.eval()
    _id2name = _load_id2name()

    print(f"✅ Model loaded and ready on {DEVICE}")
    return _model, _processor, _id2name


# ─────────────────────────────────────────────
# CLASSIFY IMAGE BYTES
# ─────────────────────────────────────────────
def classify_bytes(image_bytes: bytes):
    model, processor, id2name = load_classification_model()

    # Load and preprocess image
    image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
    inputs = processor(images=image, return_tensors="pt", padding="True").to(DEVICE)

    # Forward pass
    with torch.no_grad():
        outputs = model(**inputs)
        probs = F.softmax(outputs.logits, dim=-1)

    # Top-5 predictions
    topk = torch.topk(probs, k=5)
    indices = topk.indices[0].tolist()
    values = topk.values[0].tolist()

    results = []
    for rank, (idx, prob) in enumerate(zip(indices, values), 1):
        label = id2name.get(int(idx), f"Class {idx}")
        results.append({
            "rank": rank,
            "id": int(idx),
            "label": label,
            "score": float(prob),
        })

    # ───────────────────────────────
    # MATPLOTLIB TOP-3 PLOT
    # ───────────────────────────────
    top3 = results[:3]
    labels = [p["label"] for p in top3]
    probs_top3 = [p["score"] * 100 for p in top3]

    plt.style.use("seaborn-v0_8-whitegrid")
    fig, ax = plt.subplots(1, 2, figsize=(9, 4))

    # Left: input image
    ax[0].imshow(image)
    ax[0].axis("off")
    ax[0].set_title("Input Image", fontsize=12, weight="bold")

    # Right: horizontal bar chart
    bars = ax[1].barh(labels[::-1], probs_top3[::-1],
                      color=["#C44E52", "#55A868", "#4C72B0"],
                      edgecolor="none", height=0.6)
    ax[1].set_xlim(0, 100)
    ax[1].set_xlabel("Probability (%)", fontsize=11)
    ax[1].set_title("Top-3 Predicted Species (DINOv2-G)", fontsize=12, weight="bold")

    for bar, prob in zip(bars, probs_top3[::-1]):
        ax[1].text(prob + 1, bar.get_y() + bar.get_height()/2,
                   f"{prob:.1f}%", va="center", fontsize=10, weight="bold")

    plt.tight_layout()

    # Encode plot as base64
    buf = io.BytesIO()
    plt.savefig(buf, format="png", bbox_inches="tight")
    plt.close(fig)
    buf.seek(0)
    plot_b64 = base64.b64encode(buf.read()).decode("utf-8")
    buf.close()

    # ───────────────────────────────
    # FINAL OUTPUT
    # ───────────────────────────────
    return {
        "top1": results[0],
        "top5": results,
        "plot": f"data:image/png;base64,{plot_b64}"
    }