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

I2I Mamba

I2I Mamba is an open-source Python and PyTorch implementation of an adversarial model for multimodal medical image synthesis and image-to-image translation. It combines a convolutional backbone with channel-mixed Mamba blocks that use selective state-space layers for spatial context and channel-mixing layers for feature-channel context.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

I2I Mamba — 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 it when you have aligned medical-imaging data and want a code-based starting point for modality-translation research, rather than a hosted image-generation tool.

Who it is for

  • Medical-imaging researchers building cross-modality synthesis experiments.
  • Developers studying selective state-space components in image-to-image translation models.

Documented uses

  • Prepare aligned 2D medical-image slices for one-to-one or many-to-one translation tasks.
  • Train a model to synthesize a target modality from one or more source modalities.
  • Explore a convolutional image-translation architecture that incorporates channel-mixed Mamba blocks.
Before you choose

Requirements & limits

The public GitHub repository specifies an MIT license.

  • The documented input format requires aligned slices; corresponding images should share slice names.
  • Many-to-one tasks use two source modalities in the red and green channels, according to the README.
  • The listed environment includes CUDA 11.6 or later, PyTorch 2.2 or later, torchvision 0.17 or later and mamba-ssm 1.1.3.
  • The README separates data preparation, path-file generation and model training into distinct stages.

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 aligned images for a translation experiment

Format one-to-one or many-to-one medical-image slices using the repository's preprocessing instructions.

Before you start

  • A dataset of corresponding medical-image modalities and the repository's data_preparation scripts.
  • Python 3 and the image-processing dependencies listed in the README.
  • For many-to-one input, source modalities assigned to the red and green channels.

Steps

  1. Select subjects, skull-strip the 3D volumes, extract 2D cross-sections and normalize the slices.

  2. Arrange corresponding slices with matching names in modality folders; the README describes aligned source-and-target images placed side by side.

  3. For one-to-one preparation, set the first image path, second image path and output path in dataprep_one_to_one.py, then run it.

  4. For many-to-one preparation, set the three image paths and output path in dataprep_many_to_one.py, then run it.

Expected outcome

A prepared aligned dataset for the selected one-to-one or many-to-one task, following the README's image layout.

Guide limitations

  • The guide covers data preparation only; it does not verify that a particular dataset meets the model's needs.
  • The README specifies 512-by-256 images for one-to-one tasks and 768-by-256 for many-to-one tasks.
  • The source and target images must correspond to the same slice; the README instructs matching slice names.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
icon-lab/I2I-Mamba · Official homepage title · retained in its source language
How we review sourcesSuggest a correction