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

Dlt

DLT is a Python implementation of a layout-generation model that uses joint discrete-continuous diffusion to produce page or interface arrangements. It supports unconditional generation and conditions based on category or category plus size, with repository workflows for training and evaluation on RICO, PubLayNet, and Magazine datasets.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Dlt — 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 DLT when you need a locally run implementation for experimenting with conditioned layout generation or training and evaluating layout models on the supported datasets.

Who it is for

  • Researchers studying generative layout models.
  • Developers building or evaluating layout-generation workflows with Python and command-line tools.

Documented uses

  • Generate candidate layouts without a condition, from a category, or from category plus size.
  • Train DLT using one of the provided dataset configurations.
  • Generate test-set samples from trained weights and calculate evaluation metrics.
Before you choose

Requirements & limits

The repository is public and identifies its license as Apache-2.0.

  • This is a Python code repository and CLI workflow, not a hosted layout-generation service.
  • The README lists Ubuntu 18.04, CUDA 11.6, and Python 3.9 as its development environment and recommends Conda for dependency setup.
  • RICO, PubLayNet, and Magazine data must be obtained separately and configured locally.
  • Evaluation depends on trained weights; metric calculation also requires the FID model and a configured pickle-file path. The README describes the code as accelerator agnostic.

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

Train DLT on PubLayNet

Prepare the documented environment and start a PubLayNet training run.

Before you start

  • Ubuntu 18.04, CUDA 11.6, and Python 3.9 are the listed development environment.
  • Conda and the repository's environment.yml.
  • A downloaded PubLayNet dataset, with its location set in ./dlt/configs/remote/dataset_config.yaml.
  • A work-directory path for training output.

Steps

  1. Create the Conda environment with conda env create -f environment.yml and activate it with conda activate dlt.

  2. Download PubLayNet and update the dataset location in ./dlt/configs/remote/dataset_config.yaml.

  3. From the dlt directory, run the command below.

    Documented command
    python main.py --config configs/remote/dlt_publaynet_config.yaml --workdir

Expected outcome

The documented training workflow stores weights and related run files under the logs directory for the chosen work directory.

Guide limitations

  • The guide covers starting training; it does not establish that a run completes successfully on a particular machine.
  • The README does not specify dataset download commands or provide a fixed work-directory value.
  • The repository says the code is accelerator agnostic but lists CUDA 11.6 in its development environment.
Documentation 1 ↗
Source-based practical guide

Generate evaluation samples and calculate metrics

Use a trained model to generate test-set samples and run the documented metric workflow.

Before you start

  • A trained model and its weights placed in the configuration's logs folder.
  • A selected dataset configuration for PubLayNet, RICO, or Magazine.
  • A work-directory name and model epoch number.
  • The metric workflow requires an FID model download and the generated pickle file path to be set in dlt/evaluation/metric_comp.py.

Steps

  1. Set DATASET to publaynet, rico, or magazine in the sample-generation workflow.

  2. Run the command below, choosing all, whole_box, or loc for unconditional, category, or category-plus-size generation.

    Documented command
    python generate_samples.py --config configs/remote/dlt_{$DATASET}_config.yaml --workdir --epoch --cond_type --save True

  3. Run ./download_fid_model.sh and update the pickle-file path in dlt/evaluation/metric_comp.py.

  4. Run the command below to calculate metrics.

    Documented command
    python metric_comp.py

Expected outcome

Generated samples are saved under `logs//samples` as pickle data, which the metric script uses for evaluation.

Guide limitations

  • The README's evaluation sequence assumes the model has already been trained.
  • The documented generation and metric commands require the relevant files and paths to be configured; no result is guaranteed.
  • The source displays the DATASET assignment as DATASET = ...; adapt the placeholder in the sample command to the chosen dataset configuration.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
wix-incubator/DLT · Official homepage title · retained in its source language
How we review sourcesSuggest a correction