OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Dis

Dis is an open-source research repository for Highly Accurate Dichotomous Image Segmentation, with IS-Net code that produces image segmentation maps for foreground-background separation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Dis — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Dis when you need a documented local IS-Net inference or validation workflow for image foreground segmentation, or want research code and evaluation support for dichotomous segmentation.

Who it is for

  • Computer-vision researchers evaluating image segmentation methods.
  • Developers building image-editing or background-removal workflows who can work with a local Python and Conda implementation.

Documented uses

  • Generate predicted segmentation maps for images without ground-truth labels.
  • Run validation inference on datasets with or without ground truth.
  • Use foreground masks as an input to background removal, image editing or related visual workflows.
  • Evaluate segmentation results using the provided Human Correction Efforts metric.
Before you choose

Requirements & limits

The code and evaluation metric use the Apache License 2.0. The DIS5K dataset has separate terms of use. Repository documentation also identifies Bohrium Notebook and Hugging Face Spaces demos.

  • The repository documents Conda setup using pytorch18.yml and a PyTorch 1.8 environment.
  • Simple inference requires separately downloaded IS-Net weights and configured input and output directories.
  • The repository warns that the academic DIS release may have limited robustness for underrepresented categories, including people, animals and cars.
  • The general-use IS-Net weights are not the unreleased DIS V2.0 model or dataset.

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

Run simple inference on an image dataset

Generate predicted segmentation maps for your own images without ground truth using the repository's simple inference script.

Before you start

  • A local copy of the DIS repository.
  • Conda and the environment file at DIS/ISNet/pytorch18.yml, or dependencies configured from requirements.txt.
  • The IS-Net general-use weights, stored in saved_models/IS-Net.
  • Your input images and writable output directory.

Steps

  1. Clone the repository and create the Conda environment from DIS/ISNet/pytorch18.yml; activate the pytorch18 environment.

  2. Download isnet-general-use.pth and store it in saved_models/IS-Net.

  3. Edit ISNet/inference.py to configure the input and output directories.

  4. Run python inference.py from the ISNet directory.

Expected outcome

The inference script is configured to process the supplied images and write predicted maps to the selected output directory.

Guide limitations

  • The repository cautions that the academic DIS release may perform poorly on underrepresented targets such as people, animals and cars.
  • DIS V2.0 is distinct from the general-use IS-Net weights; the complete V2.0 dataset and model are stated as unreleased.
  • The DIS5K dataset has separate terms of use from the code and evaluation metric.
Documentation 1 ↗
Evidence and transparency

Sources & review

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