juliozhao commited on
Commit
3fa0ec1
1 Parent(s): 4a7c6e1

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +27 -172
README.md CHANGED
@@ -1,172 +1,27 @@
1
- ---
2
- license: apache-2.0
3
- dataset_info:
4
- features:
5
- - name: filename
6
- dtype: string
7
- - name: image_data
8
- dtype: binary
9
- - name: anno_string
10
- sequence: string
11
- splits:
12
- - name: train
13
- num_bytes: 3773585
14
- num_examples: 10
15
- - name: part5
16
- num_bytes: 4710765
17
- num_examples: 10
18
- - name: part6
19
- num_bytes: 4084741
20
- num_examples: 10
21
- - name: part7
22
- num_bytes: 3337016
23
- num_examples: 10
24
- - name: part8
25
- num_bytes: 4280319
26
- num_examples: 10
27
- - name: part9
28
- num_bytes: 4275308
29
- num_examples: 10
30
- - name: part10
31
- num_bytes: 3894232
32
- num_examples: 10
33
- - name: part11
34
- num_bytes: 4446364
35
- num_examples: 10
36
- - name: part12
37
- num_bytes: 3820398
38
- num_examples: 10
39
- - name: part13
40
- num_bytes: 4323211
41
- num_examples: 10
42
- - name: part14
43
- num_bytes: 3760233
44
- num_examples: 10
45
- - name: part15
46
- num_bytes: 4293369
47
- num_examples: 10
48
- - name: part16
49
- num_bytes: 4294097
50
- num_examples: 10
51
- - name: part17
52
- num_bytes: 4065835
53
- num_examples: 10
54
- - name: part18
55
- num_bytes: 3540881
56
- num_examples: 10
57
- - name: part19
58
- num_bytes: 3826798
59
- num_examples: 10
60
- - name: part20
61
- num_bytes: 4692252
62
- num_examples: 10
63
- - name: part21
64
- num_bytes: 4534566
65
- num_examples: 10
66
- - name: part22
67
- num_bytes: 4260706
68
- num_examples: 10
69
- - name: part23
70
- num_bytes: 3774230
71
- num_examples: 10
72
- - name: part24
73
- num_bytes: 4128026
74
- num_examples: 10
75
- - name: part25
76
- num_bytes: 4316172
77
- num_examples: 10
78
- - name: part26
79
- num_bytes: 3850438
80
- num_examples: 10
81
- - name: part27
82
- num_bytes: 3376316
83
- num_examples: 10
84
- - name: part28
85
- num_bytes: 4066186
86
- num_examples: 10
87
- - name: part29
88
- num_bytes: 3773585
89
- num_examples: 10
90
- - name: part1
91
- num_bytes: 4038594557
92
- num_examples: 10000
93
- - name: part2
94
- num_bytes: 4011921705
95
- num_examples: 10000
96
- - name: part3
97
- num_bytes: 4033630263
98
- num_examples: 10000
99
- - name: part4
100
- num_bytes: 4013766608
101
- num_examples: 10000
102
- - name: part0
103
- num_bytes: 4043989791
104
- num_examples: 10000
105
- download_size: 23968200707
106
- dataset_size: 20247402553
107
- configs:
108
- - config_name: default
109
- data_files:
110
- - split: train
111
- path: data/train-*
112
- - split: part0
113
- path: data/part0-*
114
- - split: part1
115
- path: data/part1-*
116
- - split: part2
117
- path: data/part2-*
118
- - split: part3
119
- path: data/part3-*
120
- - split: part4
121
- path: data/part4-*
122
- - split: part5
123
- path: data/part5-*
124
- - split: part6
125
- path: data/part6-*
126
- - split: part7
127
- path: data/part7-*
128
- - split: part8
129
- path: data/part8-*
130
- - split: part9
131
- path: data/part9-*
132
- - split: part10
133
- path: data/part10-*
134
- - split: part11
135
- path: data/part11-*
136
- - split: part12
137
- path: data/part12-*
138
- - split: part13
139
- path: data/part13-*
140
- - split: part14
141
- path: data/part14-*
142
- - split: part15
143
- path: data/part15-*
144
- - split: part16
145
- path: data/part16-*
146
- - split: part17
147
- path: data/part17-*
148
- - split: part18
149
- path: data/part18-*
150
- - split: part19
151
- path: data/part19-*
152
- - split: part20
153
- path: data/part20-*
154
- - split: part21
155
- path: data/part21-*
156
- - split: part22
157
- path: data/part22-*
158
- - split: part23
159
- path: data/part23-*
160
- - split: part24
161
- path: data/part24-*
162
- - split: part25
163
- path: data/part25-*
164
- - split: part26
165
- path: data/part26-*
166
- - split: part27
167
- path: data/part27-*
168
- - split: part28
169
- path: data/part28-*
170
- - split: part29
171
- path: data/part29-*
172
- ---
 
1
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6541f9a9eccc4f48dce3d2fb/Q-6ai0oZQpIEYduDciuhK.png)
2
+
3
+ DocSynth300K is a large-scale and diverse document layout analysis pre-training dataset, which can largely boost model performance.
4
+
5
+ ### Data Download
6
+
7
+ Use following command to download dataset(about 113G):
8
+
9
+ ```python
10
+ from huggingface_hub import snapshot_download
11
+ # Download DocSynth300K
12
+ snapshot_download(repo_id="juliozhao/DocSynth300K", local_dir="./docsynth300k-hf", repo_type="dataset")
13
+ # If the download was disrupted and the file is not complete, you can resume the download
14
+ snapshot_download(repo_id="juliozhao/DocSynth300K", local_dir="./docsynth300k-hf", repo_type="dataset", resume_download=True)
15
+ ```
16
+
17
+ ### Data Formatting & Pre-training
18
+
19
+ 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```.
20
+
21
+ ```bash
22
+ python format_docsynth300k.py
23
+ ```
24
+
25
+ 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.
26
+
27
+ **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.