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

Factualscenegraph

Factualscenegraph is a Python research benchmark and toolkit that parses text into scene graphs of entities, attributes and relations, and provides pretrained parsers, datasets and evaluation tools.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Factualscenegraph — 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 it when a research or NLP workflow needs structured entities and relations extracted from text, or when evaluating parsers on single-sentence and multi-sentence descriptions.

Who it is for

  • Researchers evaluating the faithfulness and consistency of textual scene graph parsers.
  • Developers and research teams converting captions or longer descriptions into structured entities and relations.
  • Users training or evaluating parsing models with the included benchmark datasets.

Documented uses

  • Convert a sentence into a structured scene graph.
  • Parse multi-sentence descriptions by splitting, merging and deduplicating sentence-level graphs.
  • Train or evaluate scene graph parsing models with the supplied datasets.
  • Compare parser outputs using SPICE, Soft-SPICE and Set Match.
Before you choose

Requirements & limits

The repository is public and identifies the project as open source. The README documents Python 3.8 or later and installation through pip.

  • The default mode is for single sentences; sentence_merge and DiscoSG-Refiner support multi-sentence parsing.
  • The toolkit documents CPU, CUDA and Apple Silicon MPS device choices.
  • Core parsing dependencies differ from the additional dependencies listed for refinement and evaluation.

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

Install FactualSceneGraph for a Python parsing workflow

Set up the package and its documented device prerequisites before using a pretrained parser.

Before you start

  • Python 3.8 or later.
  • Install PyTorch, Transformers, NLTK and spaCy as core dependencies.
  • For CUDA, install a CUDA-enabled PyTorch build before installing the package.

Steps

  1. Install the package with the command below.

    Documented command
    pip install FactualSceneGraph

  2. Choose the documented device setting: CPU, device='mps' for Apple Silicon, or CUDA after preparing the CUDA-enabled PyTorch installation.

Expected outcome

The package and its core dependencies are set up for a subsequent parsing workflow.

Guide limitations

  • This setup guidance does not itself run a parser or confirm a successful installation.
  • Refinement and evaluation workflows list additional dependencies.
Documentation 1 ↗
Source-based practical guide

Parse a single sentence with a pretrained model

Convert a supplied sentence into text-form scene graph relations with the documented Python parser interface.

Before you start

  • FactualSceneGraph installed in a Python 3.8-or-later environment.
  • Core dependencies: PyTorch, Transformers, NLTK and spaCy.
  • A documented device choice such as CPU.

Steps

  1. Import SceneGraphParser from factual_scene_graph.parser.scene_graph_parser.

  2. Initialize it with the pretrained model identifier lizhuang144/flan-t5-base-VG-factual-sg and a device setting.

  3. Call parse with a list containing the sentence, beam_size=1, and return_text=True.

Expected outcome

The example returns scene graph text containing entities and relations; the README's example also represents attributes as relations such as `is`.

Guide limitations

  • The documented minimal example covers single-sentence parsing; multi-sentence inputs use a separate parser mode.
  • This is repository guidance, not a compatibility guarantee or a reported execution test.
Documentation 1 ↗
Evidence and transparency

Sources & review

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