Update README.md

README.md

@@ -1,172 +1,27 @@
----
-license: apache-2.0
-dataset_info:
-  features:
-  - name: filename
-    dtype: string
-  - name: image_data
-    dtype: binary
-  - name: anno_string
-    sequence: string
-  splits:
-  - name: train
-    num_bytes: 3773585
-    num_examples: 10
-  - name: part5
-    num_bytes: 4710765
-    num_examples: 10
-  - name: part6
-    num_bytes: 4084741
-    num_examples: 10
-  - name: part7
-    num_bytes: 3337016
-    num_examples: 10
-  - name: part8
-    num_bytes: 4280319
-    num_examples: 10
-  - name: part9
-    num_bytes: 4275308
-    num_examples: 10
-  - name: part10
-    num_bytes: 3894232
-    num_examples: 10
-  - name: part11
-    num_bytes: 4446364
-    num_examples: 10
-  - name: part12
-    num_bytes: 3820398
-    num_examples: 10
-  - name: part13
-    num_bytes: 4323211
-    num_examples: 10
-  - name: part14
-    num_bytes: 3760233
-    num_examples: 10
-  - name: part15
-    num_bytes: 4293369
-    num_examples: 10
-  - name: part16
-    num_bytes: 4294097
-    num_examples: 10
-  - name: part17
-    num_bytes: 4065835
-    num_examples: 10
-  - name: part18
-    num_bytes: 3540881
-    num_examples: 10
-  - name: part19
-    num_bytes: 3826798
-    num_examples: 10
-  - name: part20
-    num_bytes: 4692252
-    num_examples: 10
-  - name: part21
-    num_bytes: 4534566
-    num_examples: 10
-  - name: part22
-    num_bytes: 4260706
-    num_examples: 10
-  - name: part23
-    num_bytes: 3774230
-    num_examples: 10
-  - name: part24
-    num_bytes: 4128026
-    num_examples: 10
-  - name: part25
-    num_bytes: 4316172
-    num_examples: 10
-  - name: part26
-    num_bytes: 3850438
-    num_examples: 10
-  - name: part27
-    num_bytes: 3376316
-    num_examples: 10
-  - name: part28
-    num_bytes: 4066186
-    num_examples: 10
-  - name: part29
-    num_bytes: 3773585
-    num_examples: 10
-  - name: part1
-    num_bytes: 4038594557
-    num_examples: 10000
-  - name: part2
-    num_bytes: 4011921705
-    num_examples: 10000
-  - name: part3
-    num_bytes: 4033630263
-    num_examples: 10000
-  - name: part4
-    num_bytes: 4013766608
-    num_examples: 10000
-  - name: part0
-    num_bytes: 4043989791
-    num_examples: 10000
-  download_size: 23968200707
-  dataset_size: 20247402553
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: part0
-    path: data/part0-*
-  - split: part1
-    path: data/part1-*
-  - split: part2
-    path: data/part2-*
-  - split: part3
-    path: data/part3-*
-  - split: part4
-    path: data/part4-*
-  - split: part5
-    path: data/part5-*
-  - split: part6
-    path: data/part6-*
-  - split: part7
-    path: data/part7-*
-  - split: part8
-    path: data/part8-*
-  - split: part9
-    path: data/part9-*
-  - split: part10
-    path: data/part10-*
-  - split: part11
-    path: data/part11-*
-  - split: part12
-    path: data/part12-*
-  - split: part13
-    path: data/part13-*
-  - split: part14
-    path: data/part14-*
-  - split: part15
-    path: data/part15-*
-  - split: part16
-    path: data/part16-*
-  - split: part17
-    path: data/part17-*
-  - split: part18
-    path: data/part18-*
-  - split: part19
-    path: data/part19-*
-  - split: part20
-    path: data/part20-*
-  - split: part21
-    path: data/part21-*
-  - split: part22
-    path: data/part22-*
-  - split: part23
-    path: data/part23-*
-  - split: part24
-    path: data/part24-*
-  - split: part25
-    path: data/part25-*
-  - split: part26
-    path: data/part26-*
-  - split: part27
-    path: data/part27-*
-  - split: part28
-    path: data/part28-*
-  - split: part29
-    path: data/part29-*
----
+![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.