from uuid import uuid4 import gradio as gr from laia.scripts.htr.decode_ctc import run as decode from laia.common.arguments import CommonArgs, DataArgs, TrainerArgs, DecodeArgs import sys from tempfile import NamedTemporaryFile, mkdtemp from pathlib import Path from contextlib import redirect_stdout import re from huggingface_hub import snapshot_download images = Path(mkdtemp()) IMAGE_ID_PATTERN = r"(?P[-a-z0-9]{36})" CONFIDENCE_PATTERN = r"(?P[0-9.]+)" # For line TEXT_PATTERN = r"\s*(?P.*)\s*" LINE_PREDICTION = re.compile(rf"{IMAGE_ID_PATTERN} {CONFIDENCE_PATTERN} {TEXT_PATTERN}") models_name = ["Teklia/pylaia-rimes"] MODELS = {} DEFAULT_HEIGHT = 128 def get_width(image, height=DEFAULT_HEIGHT): aspect_ratio = image.width / image.height return height * aspect_ratio def load_model(model_name): if model_name not in MODELS: MODELS[model_name] = Path(snapshot_download(model_name)) return MODELS[model_name] def predict(model_name, input_img): model_dir = load_model(model_name) temperature = 2.0 batch_size = 1 weights_path = model_dir / "weights.ckpt" syms_path = model_dir / "syms.txt" language_model_params = {"language_model_weight": 1.0} use_language_model = (model_dir / "tokens.txt").exists() if use_language_model: language_model_params.update( { "language_model_path": str(model_dir / "language_model.arpa.gz"), "lexicon_path": str(model_dir / "lexicon.txt"), "tokens_path": str(model_dir / "tokens.txt"), } ) common_args = CommonArgs( checkpoint=str(weights_path.relative_to(model_dir)), train_path=str(model_dir), experiment_dirname="", ) data_args = DataArgs(batch_size=batch_size, color_mode="L") trainer_args = TrainerArgs( # Disable progress bar else it messes with frontend display progress_bar_refresh_rate=0 ) decode_args = DecodeArgs( include_img_ids=True, join_string="", convert_spaces=True, print_line_confidence_scores=True, print_word_confidence_scores=False, temperature=temperature, use_language_model=use_language_model, **language_model_params, ) with NamedTemporaryFile() as pred_stdout, NamedTemporaryFile() as img_list: image_id = uuid4() # Resize image to 128 if bigger/smaller input_img = input_img.resize((int(get_width(input_img)), DEFAULT_HEIGHT)) input_img.save(str(images / f"{image_id}.jpg")) # Export image list Path(img_list.name).write_text("\n".join([str(image_id)])) # Capture stdout as that's where PyLaia outputs predictions with redirect_stdout(open(pred_stdout.name, mode="w")): decode( syms=str(syms_path), img_list=img_list.name, img_dirs=[str(images)], common=common_args, data=data_args, trainer=trainer_args, decode=decode_args, num_workers=1, ) # Flush stdout to avoid output buffering sys.stdout.flush() predictions = Path(pred_stdout.name).read_text().strip().splitlines() assert len(predictions) == 1 _, score, text = LINE_PREDICTION.match(predictions[0]).groups() return input_img, {"text": text, "score": score} gradio_app = gr.Interface( predict, inputs=[ gr.Dropdown(models_name, value=models_name[0], label="Models"), gr.Image( label="Upload an image of a line", sources=["upload", "clipboard"], type="pil", height=DEFAULT_HEIGHT, width=2000, image_mode="L", ), ], outputs=[ gr.Image(label="Processed Image"), gr.JSON(label="Decoded text"), ], examples=[ ["Teklia/pylaia-rimes", str(filename)] for filename in Path("examples").iterdir() ], title="Decode the transcription of an image using a PyLaia model", cache_examples=True, ) if __name__ == "__main__": gradio_app.launch()