| """NotavelAI: identify Brazilian Real banknotes in photos / identifica cédulas de Real em fotos. |
| |
| Usage / uso: |
| pip install tensorflow pillow numpy |
| python predict.py photo.jpg [other.jpg ...] |
| |
| Runs the model exactly like the Android app: the central square of the photo, resized to 224x224, |
| and a value is only announced with at least 70% confidence. |
| Roda o modelo como o app Android: o quadrado central da foto, reduzido a 224x224, e só anuncia |
| um valor com pelo menos 70% de confiança. |
| """ |
|
|
| import sys |
| from pathlib import Path |
|
|
| import numpy as np |
| import tensorflow as tf |
| from PIL import Image, ImageOps |
|
|
| HERE = Path(__file__).resolve().parent |
| MODEL = HERE / "tflite" / "notavelai.tflite" |
| LABELS = (HERE / "labels.txt").read_text(encoding="utf-8").split() |
| THRESHOLD = 0.7 |
| SIZE = 224 |
|
|
|
|
| def load(path: str) -> np.ndarray: |
| """Fix EXIF rotation, cap the longer side at 1024 px (like the app), take the central square |
| and resize it to 224x224 with bilinear interpolation. Values stay 0-255 (not normalized).""" |
| with Image.open(path) as image: |
| rgb = ImageOps.exif_transpose(image).convert("RGB") |
| rgb.thumbnail((1024, 1024)) |
| array = np.asarray(rgb, dtype="float32") |
| height, width = array.shape[:2] |
| side = min(height, width) |
| top, left = (height - side) // 2, (width - side) // 2 |
| square = array[top:top + side, left:left + side] |
| return tf.image.resize(square, (SIZE, SIZE), method="bilinear").numpy()[None] |
|
|
|
|
| def run(image: np.ndarray) -> np.ndarray: |
| interpreter = tf.lite.Interpreter(model_path=str(MODEL)) |
| interpreter.allocate_tensors() |
| interpreter.set_tensor(interpreter.get_input_details()[0]["index"], image) |
| interpreter.invoke() |
| return interpreter.get_tensor(interpreter.get_output_details()[0]["index"])[0] |
|
|
|
|
| def main() -> None: |
| if len(sys.argv) < 2: |
| sys.exit(__doc__) |
| for path in sys.argv[1:]: |
| probabilities = run(load(path)) |
| best = int(np.argmax(probabilities)) |
| if LABELS[best] == "fundo": |
| answer = "no banknote / nenhuma cédula" |
| elif probabilities[best] < THRESHOLD: |
| answer = "not sure / não consegui identificar" |
| else: |
| answer = f"R$ {LABELS[best]}" |
| print(f"{path}: {answer} ({probabilities[best]:.0%})") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
