DECIMER can not recognized SMILES length larger than 302

[wangtaiyi]

Issue Type

Bug

Source

GitHub (source)

DECIMER Image Transformer Version

2.8

OS Platform and Distribution

Ubuntu 22.04

Python version

3.10

Current Behaviour?

I am a researcher on natural products, but sometimes the molecules are quite large, like MW > 1000. When I tried to OCR some molecules from images, I found more than 5% of molecules can not be OCRed into full length SMILES. Is it possible to enlarge the max_length of SMILES?

Many thanks!!

Which images caused the issue? (This is mandatory for images related issues)

Standalone code to reproduce the issue

#!/usr/bin/env python3
"""
decimer_batch.py
断点续跑：先建表，再逐行回填 SMILES
支持从 config.yaml 读取输入/输出路径
"""

import os
import glob
import yaml
import pandas as pd
from tqdm import tqdm
from DECIMER import predict_SMILES   # pip install decimer

# 支持的图片后缀
IMG_EXTS = ("*.png", "*.jpg", "*.jpeg", "*.tif", "*.tiff", "*.bmp")

def gather_image_paths(folder):
    """返回所有图片的绝对路径列表"""
    paths = []
    for ext in IMG_EXTS:
        paths.extend(glob.glob(os.path.join(folder, ext)))
    return sorted(paths)

def init_csv(folder, csv_file):
    """第一次运行时建立仅有 image_path 的 csv"""
    paths = gather_image_paths(folder)
    if not paths:
        raise RuntimeError("未找到任何图片！")
    df = pd.DataFrame({
        "row_id": range(1, len(paths) + 1),
        "image_path": [os.path.abspath(p) for p in paths],
        "smiles": [None] * len(paths)
    })
    os.makedirs(os.path.dirname(csv_file), exist_ok=True)
    df.to_csv(csv_file, index=False)
    print(f"已创建 {csv_file}，共 {len(paths)} 条待处理记录。")
    return df

def resume_convert(csv_file):
    """读取 csv，找到未完成的行继续转换"""
    df = pd.read_csv(csv_file)

    mask_na = df["smiles"].isna()
    todo = df[mask_na]

    if todo.empty:
        print("所有图片均已转换完成。")
        return

    print(f"发现 {len(todo)} 张图片未完成，继续转换……")

    for idx in tqdm(todo.index, desc="DECIMER"):
        img_path = df.at[idx, "image_path"]
        print(img_path)
        try:
            smiles = predict_SMILES(img_path).strip()
            if not smiles:
                smiles = None
        except Exception as e:
            print(f"[WARN] 处理 {img_path} 出错：{e}")
            smiles = None

        # 回填并立即保存
        df.at[idx, "smiles"] = smiles
        df.to_csv(csv_file, index=False)

    print("全部完成！")

def main():
    import argparse
    parser = argparse.ArgumentParser(description="可断点续跑的批量图片→SMILES（支持 config.yaml）")
    parser.add_argument("-c", "--config", default="config.yaml", help="配置文件路径 (default: config.yaml)")
    parser.add_argument("-o", "--out", help="输出 CSV 文件名（可选，覆盖配置中的默认值）")
    args = parser.parse_args()

    # 读取 config.yaml
    with open(args.config, 'r', encoding='utf-8') as f:
        config = yaml.safe_load(f)

    output_dir = config['output']['output_dir']
    img_folder = os.path.join(output_dir, "imgs")

    # 确定 CSV 输出路径
    if args.out:
        csv_file = args.out
    else:
        csv_file = os.path.join(output_dir, "results.csv")

    if not os.path.isfile(csv_file):
        # 第一次运行：先建表
        init_csv(img_folder, csv_file)

    # 无论新建还是继续，都执行 resume_convert
    resume_convert(csv_file)

if __name__ == "__main__":
    main()

Relevant log output

Code of Conduct

• I agree to follow this project's Code of Conduct