![image/png](https://cdn-uploads.huggingface.co/production/uploads/6541f9a9eccc4f48dce3d2fb/Q-6ai0oZQpIEYduDciuhK.png)

DocSynth300K is a large-scale and diverse document layout analysis pre-training dataset, which can largely boost model performance.

### Data Download

Use following command to download dataset(about 113G):

```python
from huggingface_hub import snapshot_download
# Download DocSynth300K
snapshot_download(repo_id="juliozhao/DocSynth300K", local_dir="./docsynth300k-hf", repo_type="dataset")
# If the download was disrupted and the file is not complete, you can resume the download
snapshot_download(repo_id="juliozhao/DocSynth300K", local_dir="./docsynth300k-hf", repo_type="dataset", resume_download=True)
```

### Data Formatting & Pre-training

If you want to perform DocSynth300K pretraining, using ```format_docsynth300k.py``` to convert original ```.parquet``` format into ```YOLO``` format. The converted data will be stored at ```./layout_data/docsynth300k```.

```bash
python format_docsynth300k.py
```

To perform DocSynth300K pre-training, use this [command](assets/script.sh#L2). We default use 8GPUs to perform pretraining. To reach optimal performance, you can adjust hyper-parameters such as ```imgsz```, ```lr``` according to your downstream fine-tuning data distribution or setting.

**Note:** Due to memory leakage in YOLO original data loading code, the pretraining on large-scale dataset may be interrupted unexpectedly, use ```--pretrain last_checkpoint.pt --resume``` to resume the pretraining process.

https://huggingface.co/papers/2410.12628