PaddlePaddle/PaddleOCR · error · Exception

{} does not exist!

Error message

{} does not exist!

What it means

The PubTabNet table-recognition dataset reads JSONL annotations (filename, html.cells, html.structure.tokens), joins data_dir with info['filename'], and raises Exception('{path} does not exist!') when the table image is absent. The bare except then logs and skips the sample.

Source

Thrown at ppocr/data/pubtab_dataset.py:106

    def shuffle_data_random(self):
        if self.do_shuffle:
            random.seed(self.seed)
            random.shuffle(self.data_lines)
        return

    def __getitem__(self, idx):
        try:
            data_line = self.data_lines[idx]
            data_line = data_line.decode("utf-8").strip("\n")
            info = json.loads(data_line)
            file_name = info["filename"]
            cells = info["html"]["cells"].copy()
            structure = info["html"]["structure"]["tokens"].copy()

            img_path = os.path.join(self.data_dir, file_name)
            if not os.path.exists(img_path):
                raise Exception("{} does not exist!".format(img_path))
            data = {
                "img_path": img_path,
                "cells": cells,
                "structure": structure,
                "file_name": file_name,
            }

            with open(data["img_path"], "rb") as f:
                img = f.read()
                data["image"] = img
            outs = transform(data, self.ops)
        except:
            import traceback

            err = traceback.format_exc()
            self.logger.error(
                "When parsing line {}, error happened with msg: {}".format(
                    data_line, err

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Verify one entry: python -c 'import json,os; i=json.loads(open("train.jsonl").readline()); print(os.path.exists(os.path.join(data_dir, i["filename"])))'
2. Set data_dir in the table yml to the directory containing the images referenced by filename
3. Download/restore the missing image splits from the PubTabNet distribution
4. If the layout differs, rewrite the JSONL filenames with a small script instead of moving gigabytes of images

Example fix

# before (config)
data_dir: ./train_data/pubtabnet
# after (images live one level down)
data_dir: ./train_data/pubtabnet/train_images

import json, os
for line in open(jsonl_path, encoding='utf-8'):
    fn = json.loads(line)['filename']
    if not os.path.exists(os.path.join(data_dir, fn)):
        raise SystemExit(f'{fn} missing under {data_dir}')

try:
    out = dataset[idx]
except Exception:
    out = None  # logged and skipped upstream; monitor skip rate

When it happens

Trigger: Training/evaluating PP-Structure table models with a PubTabNet JSONL whose 'filename' entries do not resolve under data_dir: images not downloaded, data_dir wrong, or filename containing a subdirectory that was flattened on unzip.

Common situations: Downloading PubTabNet JSONL but not the image archive; using val.jsonl with train images only; data_dir pointing at the JSONL folder instead of the image folder.

Related errors

• {} does not exist!
• {} does not exist!
• Missing LMDB dataset value
• RecResizeImg.image_shape is required in rec inference.yml
• Unexpected recognition channels: ${String(channels)}

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/8963658006caeefe. Report an issue: GitHub.