Nix-runnable version of https://layout-parser.readthedocs.io/en/latest/example/deep_layout_parsing/index.html

flake.lock

{
  "nodes": {
    "nixpkgs": {
      "locked": {
        "lastModified": 1751477932,
        "narHash": "sha256-mRbVq/Ht0Sr0kzt9wQ9IyXFIpeFz6DhZ+6if/HOcMFw=",
        "owner": "NixOS",
        "repo": "nixpkgs",
        "rev": "51e8f407be7d7ba35321d9b43acf31fc25fe7cf1",
        "type": "github"
      },
      "original": {
        "owner": "NixOS",
        "repo": "nixpkgs",
        "type": "github"
      }
    },
    "root": {
      "inputs": {
        "nixpkgs": "nixpkgs"
      }
    }
  },
  "root": "root",
  "version": 7
}

flake.nix

{
  description = "nix-runnable version of layoutparser image-to-text";

  inputs.nixpkgs.url = "github:NixOS/nixpkgs";

  outputs =
    { self, nixpkgs }:
    let
      system = "x86_64-linux";
      pkgs = nixpkgs.legacyPackages.${system};
      pythonEnv = pkgs.python312.withPackages (
        ps: with ps; [
          opencv4
          layoutparser
          detectron2
          torchvision
          pytesseract
        ]
      );
    in
    {
      devShells.${system}.default = pkgs.mkShell {
        buildInputs = [ pythonEnv ];
      };

      packages.${system}.img2txt = pkgs.writeShellApplication {
        name = "img2txt";
        runtimeInputs = [ pythonEnv ];
        text = ''
          exec ${pythonEnv.interpreter} ${self}/img2txt.py "$@"
        '';
      };

      apps.${system}.img2txt = {
        type = "app";
        program = "${self.packages.${system}.img2txt}/bin/img2txt";
      };
    };
}

img2txt.py

#!/usr/bin/env python

import argparse
import layoutparser as lp
import cv2

parser = argparse.ArgumentParser(description="Run layoutparser model on an image and extract text blocks.")
parser.add_argument("input", help="Path to the input image file")
parser.add_argument("--debug", action="store_true", help="Enable debug output and save intermediate image")
args = parser.parse_args()

# This program is largely the example from the documentation:
# https://layout-parser.readthedocs.io/en/latest/example/deep_layout_parsing/index.html

image = cv2.imread(args.input)
model = lp.Detectron2LayoutModel('lp://PubLayNet/faster_rcnn_R_50_FPN_3x/config',
                                 extra_config=["MODEL.ROI_HEADS.SCORE_THRESH_TEST", 0.8],
                                 label_map={0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"})
layout = model.detect(image)

if args.debug:
    lp.draw_box(image, layout, box_width=3).save("/tmp/debug.jpg")

text_blocks = lp.Layout([b for b in layout if b.type=='Text'])

ocr_agent = lp.TesseractAgent(languages='deu')

for block in text_blocks:
    segment_image = (block
                       .pad(left=5, right=5, top=5, bottom=5)
                       .crop_image(image))
        # add padding in each image segment can help
        # improve robustness

    text = ocr_agent.detect(segment_image)
    block.set(text=text, inplace=True)

for txt in text_blocks.get_texts():
    print(txt, end='\n---\n')