It's okay, solved the issue. I didn't put enough training data into the
folder hence no training data
On Monday, 28 April 2025 at 10:43:05 UTC+8 Jiansen Chan wrote:
> My goal is to automate model training in tesseract OCR for Japanese words.
> The user should just paste ground truth files and picture files into a
> particular folder, and then use that data to train a new model. this
> process should be able to be carried out multiple times. Every single time
> data is added to the folder I expect an automated model training.
>
> However, this is the error that i run into when I try to run automated
> tesseract training on VSCode. What I did is that I had a script that uses
> watchdog to detect newly added .tif/.png files alongside their
> corresponding .gt.txt files into a particular folder (from which the model
> is supposed to treat as training data and use it to train). The watcher
> file looks something like this:
>
> (watcher_trainng.py)
> import time
> import os
> from watchdog.observers import Observer
> from watchdog.events import FileSystemEventHandler
> from pathlib import Path
> from training.tesseract_training import run_tesseract_training
> from training.training_model_utils import get_latest_and_next_model
> WATCHED_FOLDER = r"C:\Users\Chan Jian Sen\Documents\ocr-japanese\I
> NPUT_TRAINING_DATA" #ground truth put here
> tesstrain_dir = r"C:\Users\Chan Jian Sen\Documents\TesseractFineTuningJpn5
> \tesstrain"
>
>
> class TrainingInputHandler(FileSystemEventHandler):
>
> def on_modified(self, event):
> self.check_and_trigger_training()
>
> def on_created(self, event):
> self.check_and_trigger_training()
>
> def check_and_trigger_training(self):
> files = os.listdir(WATCHED_FOLDER)
> pngs = {Path(f).stem for f in files if f.endswith('.png')}
> gts = {Path(f).stem for f in files if f.endswith('.gt.txt')}
> common = pngs & gts
>
> if len(common) == 0:
> print("⏳ Waiting for matching .png and .gt.txt pairs...")
>
> tessdata_path = r"C:\Users\Chan Jian Sen\Documents\T
> esseractFineTuningJpn5\tessdata"
> start_model, new_model = get_latest_and_next_model(tessdata_path)
>
> print(f"🔁 Using {start_model} as base, training new model: {
> new_model}") #problem here is the the old model they saw it as jpn and
> the new model as jpn1
>
> run_tesseract_training(tesstrain_dir, new_model, start_model) #the
> first parameter MUST be your tesstrain folder
> observer.stop()
>
> if __name__ == "__main__":
> print(f"👀 Watching training data folder: {WATCHED_FOLDER}")
> event_handler = TrainingInputHandler()
> observer = Observer()
> observer.schedule(event_handler, WATCHED_FOLDER, recursive=False)
> observer.start()
>
> try:
> while observer.is_alive():
> time.sleep(1)
> except KeyboardInterrupt:
> observer.stop()
> observer.join()
>
> To generate a new model name (since I want to automate model training), i
> also have these functions here:
> (training_model_utils.py)
> import os
>
> def get_model_names(tessdata_path, model_prefix="jpn"):
> models = []
> for fname in os.listdir(tessdata_path):
> if fname.startswith(model_prefix) and fname.endswith(
> ".traineddata"):
> suffix = fname[len(model_prefix):-len(".traineddata")]
> if suffix == "":
> models.append((0, "jpn"))
> elif suffix.isdigit():
> models.append((int(suffix), f"{model_prefix}{suffix}"))
> models.sort()
> return models
>
> def get_latest_and_next_model(tessdata_path, model_prefix="jpn"):
> models = get_model_names(tessdata_path, model_prefix)
> if not models:
> return model_prefix, f"{model_prefix}2"
> latest = models[-1][1]
> next_num = models[-1][0] + 1
> next_model = f"{model_prefix}{next_num}" if next_num > 0 else f"{
> model_prefix}2"
> return latest, next_model
>
>
>
>
> I also coded the make training procedure into VSCode, with a python script
> that calls for it. This code snippet below is meant to run the tesseract
> training.
> (tesseract_training.py)
> import subprocess
> import os
>
> def run_tesseract_training(training_dir, model_name, start_model,
> max_iterations=4000): #previously start model is jpn
> """
> Run the full Tesseract tesstrain workflow including unicharset and
> langdata.
> """
> tessdata_path = r"C:\Users\Chan Jian Sen\Documents\T
> esseractFineTuningJpn5\tessdata"
> # Important: replace backslashes with forward slashes
> tessdata_path = tessdata_path.replace("\\", "/")
> env = os.environ.copy()
> env["TESSDATA_PREFIX"] = tessdata_path
> command = [
> "make",
> "unicharset", "lists", "proto-model", "tesseract-langdata",
> "training",
> f"MODEL_NAME={model_name}",
> f"START_MODEL={start_model}",
> f"TESSDATA={tessdata_path}", # Adjust path depending on where
> your .traineddata are
> f"GROUND_TRUTH_DIR={training_dir}",
> f"MAX_ITERATIONS={max_iterations}",
> "LEARNING_RATE=0.001"
> ]
>
> print("🚀 Running full Tesseract training pipeline...")
> try:
> subprocess.run(command, cwd=r"C:\Users\Chan Jian Sen\Documents\T
> esseractFineTuningJpn5\tesstrain", shell=True, check=True, env=env)
> print(f"✅ Training complete: {model_name}.traineddata generated.")
> except subprocess.CalledProcessError as e:
> print(f"❌ Training failed: {e}")
>
>
> However this is my terminal output when I run the watcher file.
> PS C:\Users\Chan Jian Sen\Documents\ocr-japanese> c:; cd 'c:\Users\Chan
> Jian Sen\Documents\ocr-japanese'; & 'c:\Users\Chan Jian
> Sen\AppData\Local\Programs\Python\Python39\python.exe' 'c:\Users\Chan Jian
> Sen\.vscode\extensions\ms-python.debugpy-2025.6.0-win32-x64\bundled\libs\debugpy\launcher'
>
> '56444' '--' 'C:\Users\Chan Jian
> Sen\Documents\ocr-japanese\watcher_training.py'
> gpy-2025.6.0-win32-x64\x5cbundled\x5clibs\x5cdebugpy\x5clauncher' '56444'
> '--' 'C:\x5cUsers\x5cChan Jian
> Sen\x5cDocuments\x5cocr-japanese\x5cwatcher_training.py'
> ;0a5d0c8e-f6f4-44db-b1ea-a49791670afe👀 Watching training data folder:
> C:\Users\Chan Jian Sen\Documents\ocr-japanese\INPUT_TRAINING_DATA
>
> ⏳ Waiting for matching .png and .gt.txt pairs...
> 🔁 Using jpn as base, training new model: jpn1
> 🚀 Running full Tesseract training pipeline...
> You are using make version: 4.4.1
> Makefile:438: *** mixed implicit and normal rules: deprecated syntax
> combine_tessdata -u C:/Users/Chan Jian
> Sen/Documents/TesseractFineTuningJpn5/tessdata/jpn.traineddata data/jpn/jpn1
> Failed to read C:/Users/Chan
> make: *** [Makefile:207: data/jpn/jpn1.lstm-unicharset] Error 1
>
>
> Would greatly appreciate for any help given. Sorry if it's quite a lot to
> digest.
>
--
You received this message because you are subscribed to the Google Groups
"tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to tesseract-ocr+unsubscr...@googlegroups.com.
To view this discussion visit
https://groups.google.com/d/msgid/tesseract-ocr/eaf6bc12-bcca-4ee4-b3d6-0ce684537fe6n%40googlegroups.com.
