OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Docsegtr

Docsegtr is a Python and PyTorch implementation of a transformer-based instance-segmentation model for document images, with a stated focus on scientific document layouts and the PubLayNet benchmark. It combines CNN/FPN multiscale features, positional embeddings and transformer layers to predict instances and generate segmentation kernels. The repository documents training, evaluation and prediction visualization wor

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Docsegtr — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Docsegtr when you need code for document-layout instance segmentation and are prepared to work with a Python/PyTorch research stack, including building its framework dependencies.

Who it is for

  • Researchers studying document-image segmentation.
  • Computer-vision developers who want to train or evaluate a document-layout instance-segmentation model.
  • Teams that need to inspect qualitative layout predictions on PubLayNet.

Documented uses

  • Segment distinct layout instances in scientific document images.
  • Train the model from scratch with the repository’s provided configuration.
  • Evaluate a model using pretrained weights.
  • Visualize qualitative PubLayNet predictions from a model-generated JSON file.
Before you choose

Requirements & limits

The GitHub repository is public and lists the GPL-3.0 license.

  • The project is a model implementation that you run and develop with, not a hosted application.
  • Setup requires Conda and source builds of Detectron2 v0.2.1 and AdelaiDet.
  • The repository documents distinct evaluation, training and visualization workflows; visualization requires JSON produced by a trained model.
  • The documented benchmark path uses PubLayNet, while the model zoo also lists pretrained weights for PRIMA, HJ and Table datasets.
  • The repository identifies GPL-3.0 as its license.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the documented Docsegtr environment

Clone the repository, create its Conda environment and build the two framework dependencies required by its setup instructions.

Before you start

  • Git and Conda
  • Python environment dependencies listed in the repository’s environment.yml
  • Detectron2 v0.2.1 source and AdelaiDet source

Steps

  1. Clone https://github.com/biswassanket/DocSegTr.git and change into the repository directory.

  2. Create and activate the environment with conda env create -f environment.yml and conda activate instaseg.

  3. Build Detectron2 v0.2.1 from source using the command below in its source directory.

    Documented command
    python setup.py build develop

  4. Build AdelaiDet from source using the command below from the repository work directory, as described in the README.

    Documented command
    python setup.py build develop

Expected outcome

The repository’s Conda environment is prepared and the documented framework build steps are completed.

Guide limitations

  • This covers setup only; it does not establish that the environment builds successfully on a particular machine.
  • The README does not specify hardware requirements or provide a complete troubleshooting procedure.
Documentation 1 ↗
Source-based practical guide

Visualize PubLayNet predictions

Use the repository’s documented visualization script with an existing model-generated JSON file.

Before you start

  • A JSON file created by a trained model
  • The Docsegtr repository environment and PubLayNet visualization data

Steps

  1. Provide the JSON file path to tools/visualize_publaynet.py with the --input option.

  2. Set an output directory with --output.

  3. Set --dataset publaynet_minival and the documented --conf-threshold 0.6 option.

Expected outcome

The documented script is configured to write qualitative prediction visualizations to the selected output directory.

Guide limitations

  • The README does not explain how to create the input JSON in this visualization step.
  • The guide reflects the documented PubLayNet workflow and does not establish behavior on other datasets.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
biswassanket/DocSegTr · Official homepage title · retained in its source language
How we review sourcesSuggest a correction