image_size list | blocks list |
|---|---|
[
773,
1000
] | [
{
"class": "header",
"bbox": [
279.8,
57,
473.1,
70
],
"content": "Mutual Information Neural Estimation"
},
{
"class": "image",
"bbox": [
82.7,
82,
272.9,
178
],
"content": null
},
{
"class": "image_caption",
"bbox": [
... |
DocBank fixture
Single-page fixture with a 2-column scientific paper containing text, equations, a table, and figures. Block-level GT generated by MinerU2.5-Pro.
Source
- Image: page 6 of Mutual Information Neural Estimation (Belghazi et al., ICML 2018), from the DocBank dataset.
- Annotations:
opendatalab/MinerU2.5-Pro-2604-1.2Bviatwo_step_extract.
Usage
from huggingface_hub import hf_hub_download
import json
img = hf_hub_download("thisisiron/hf-docbank-fixture", "sample.jpg", repo_type="dataset")
gt = json.loads(open(hf_hub_download("thisisiron/hf-docbank-fixture", "gt_blocks.json", repo_type="dataset")).read())
for i, b in enumerate(gt["blocks"]):
print(i, b["class"], b["bbox"])
- Downloads last month
- 34