wendlerc/RenderedText · Datasets at Hugging Face (original) (raw)

This dataset has been created by Stability AI and LAION.

This dataset contains 12 million 1024x1024 images of handwritten text written on a digital 3D sheet of paper generated using Blender geometry nodes and rendered using Blender Cycles. The text has varying font size, color, and rotation, and the paper was rendered under random lighting conditions. Note that, the first 10 million examples are in the root folder of this dataset repository and the remaining 2 million are in ./remaining (due to the constraint on number of files per directory).

It was generated with the script https://github.com/GbotHQ/ocr-dataset-rendering/, which utilizes:

Line level annotations Character level annotations

The dataset contains both line-level, as well as character level annotations for each example. The annotations are stored in the accompanying json files and are of the following form:

{
'ocr_annotation':
{'bounding_boxes': [[[145.0, 370.0], [788.0, 353.0], [827.0, 633.0], [182.0, 669.0]]], 
 'text': ['Joe.'], 
 'bb_relative': [[[0.1416015625, 0.361328125], [0.76953125, 0.3447265625], [0.8076171875, 0.6181640625], [0.177734375, 0.6533203125]]], 
 'char': ['J', 'o', 'e', '.'], 
 'char_idx': [0, 1, 2, 3], 
 'bb_character_level': [[[145.0, 370.0], [346.0, 365.0], [382.0, 651.0], [181.0, 662.0]], [[375.0, 438.0], [557.0, 431.0], [585.0, 640.0], [402.0, 650.0]], [[578.0, 440.0], [744.0, 434.0], [771.0, 629.0], [604.0, 638.0]], [[778.0, 591.0], [821.0, 589.0], [827.0, 633.0], [784.0, 635.0]]], 
 'font_path': '/fsx/home-wendlerc/blender-dataset/assets/fonts/fontcollection/HelloScribbles-axapm.ttf', 
 'font_color': [17, 25, 231], 
 'text_rotation_angle': 7},
'width':1024,
'height':1024,
}

Browse a few more examples here: https://colab.research.google.com/drive/1o0rZhtY9aeurzNrAbu6nJypULSIIcf1v?usp=sharing

Downloads last month

29,564

Models trained or fine-tuned on wendlerc/RenderedText