microsoft / kosmos-2.5

Kosmos-2.5

Microsoft Document AI | GitHub

Model description

Kosmos-2.5 is a multimodal literate model for machine reading of text-intensive images. Pre-trained on large-scale text-intensive images, Kosmos-2.5 excels in two distinct yet cooperative transcription tasks: (1) generating spatially-aware text blocks, where each block of text is assigned its spatial coordinates within the image, and (2) producing structured text output that captures styles and structures into the markdown format. This unified multimodal literate capability is achieved through a shared decoder-only auto-regressive Transformer architecture, task-specific prompts, and flexible text representations. We evaluate Kosmos-2.5 on end-to-end document-level text recognition and image-to-markdown text generation. Furthermore, the model can be readily adapted for any text-intensive image understanding task with different prompts through supervised fine-tuning, making it a general-purpose tool for real-world applications involving text-rich images. This work also paves the way for the future scaling of multimodal large language models.

Kosmos-2.5: A Multimodal Literate Model

NOTE:

Since this is a generative model, there is a risk of hallucination during the generation process, and it CAN NOT guarantee the accuracy of all OCR/Markdown results in the images.

Use with transformers：

from PIL import Image
import requests
import torch
from transformers import AutoProcessor, Kosmos2_5ForConditionalGeneration
import re
repo = "microsoft/kosmos-2.5"
device = "cuda:0"
dtype = torch.bfloat16
model = Kosmos2_5ForConditionalGeneration.from_pretrained(repo, device_map=device, torch_dtype=dtype)
processor = AutoProcessor.from_pretrained(repo)
url = "https://huggingface.co/kirp/kosmos2_5/resolve/main/receipt_00008.png"
image = Image.open(requests.get(url, stream=True).raw)
prompt = "<ocr>" # <md>
inputs = processor(text=prompt, images=image, return_tensors="pt")
height, width = inputs.pop("height"), inputs.pop("width")
raw_width, raw_height = image.size
scale_height = raw_height / height
scale_width = raw_width / width
inputs = {k: v.to(device) if v is not None else None for k, v in inputs.items()}
inputs["flattened_patches"] = inputs["flattened_patches"].to(dtype)
generated_ids = model.generate(
    **inputs,
    max_new_tokens=1024,
)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
def postprocess(y, scale_height, scale_width):
    y = y.replace(prompt, "")
    if "<md>" in prompt:
        return y
    pattern = r"<bbox><x_\d+><y_\d+><x_\d+><y_\d+></bbox>"
    bboxs_raw = re.findall(pattern, y)
    lines = re.split(pattern, y)[1:]
    bboxs = [re.findall(r"\d+", i) for i in bboxs_raw]
    bboxs = [[int(j) for j in i] for i in bboxs]
    info = ""
    for i in range(len(lines)):
        box = bboxs[i]
        x0, y0, x1, y1 = box
        if not (x0 >= x1 or y0 >= y1):
            x0 = int(x0 * scale_width)
            y0 = int(y0 * scale_height)
            x1 = int(x1 * scale_width)
            y1 = int(y1 * scale_height)
            info += f"{x0},{y0},{x1},{y0},{x1},{y1},{x0},{y1},{lines[i]}"
    return info
output_text = postprocess(generated_text[0], scale_height, scale_width)
print(output_text)

55,595,71,595,71,629,55,629,1
82,595,481,595,481,635,82,635,[REG] BLACK SAKURA
716,590,841,590,841,629,716,629,45,455
55,637,71,637,71,672,55,672,1
82,637,486,637,486,675,82,675,COOKIE DOH SAUCES
818,632,843,632,843,668,818,668,0
51,683,71,683,71,719,51,719,1
82,683,371,683,371,719,82,719,NATA DE COCO
820,677,845,677,845,713,820,713,0
32,770,851,770,851,811,32,811,Sub Total 45,455
28,811,853,811,853,858,28,858,PB1 (10%) 4,545
28,857,855,857,855,905,28,905,Rounding 0
24,905,858,905,858,956,24,956,Total 50,000
17,1096,868,1096,868,1150,17,1150,Card Payment 50,000

Citation

If you find Kosmos-2.5 useful in your research, please cite the following paper:

@article{lv2023kosmos,
  title={Kosmos-2.5: A multimodal literate model},
  author={Lv, Tengchao and Huang, Yupan and Chen, Jingye and Cui, Lei and Ma, Shuming and Chang, Yaoyao and Huang, Shaohan and Wang, Wenhui and Dong, Li and Luo, Weiyao and others},
  journal={arXiv preprint arXiv:2309.11419},
  year={2023}
}

License

The content of this project itself is licensed under the MIT

Microsoft Open Source Code of Conduct