Practical overview
When it fits
Use it when working on research or benchmarking workflows for pansharpening and the named hyperspectral datasets, rather than when seeking a hosted image-processing application.
Who it is for
- Researchers studying pansharpening and multispectral or hyperspectral image fusion.
- Practitioners seeking a documented research workflow for training and evaluating a pansharpening model.
Documented uses
- Prepare named hyperspectral datasets into patches for model experiments.
- Pre-train the model backbone and fine-tune its MHFA modules.
- Consult linked trained models and test-set prediction outputs for the documented datasets.
Requirements & limits
The repository is public and licensed under MIT. Its README provides Conda and requirements-file environment setup instructions.
- The documented training and data-preparation workflow is organized around Pavia Center, Botswana and Chikusei.
- Dataset-specific MATLAB scripts prepare patches before the Python training stages.
- The README links to trained models and test-set prediction outputs for each of the three datasets.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the documented Conda environment
Create and activate a project environment from the repository's supplied environment file.
Before you start
- The HyperTransformer repository
- Conda
Steps
Create an environment using the repository's environment.yaml file: conda env create --name HyperTransformer --file environment.yaml.
Activate it with conda activate HyperTransformer.
Expected outcome
A Conda environment named HyperTransformer is created and activated from the supplied environment specification.
Guide limitations
- This covers environment setup only; it does not run training or verify the environment on a particular machine.
Prepare and train on a documented dataset
Follow the README's dataset-specific process and two-stage training sequence for one of the named hyperspectral datasets.
Before you start
- The repository's Python environment
- One of the Pavia Center, Botswana or Chikusei datasets placed at the repository path specified in the README
- Dataset-processing scripts and training configuration files from the repository
Steps
Place the selected dataset file at its documented path under ./datasets.
Run the matching process_pavia.m, process_botswana.m or process_chikusei.m script in its dataset directory to generate patches using the described protocol.
Set train_dataset in configs/config_HSIT_PRE.json to the selected dataset and run python train.py --config configs/config_HSIT_PRE.json.
Set train_dataset in configs/config_HSIT.json for that dataset and run python train.py with --resume pointing to the backbone checkpoint from the previous stage.
Expected outcome
The documented workflow prepares dataset patches, pre-trains the backbone and proceeds to MHFA-module fine-tuning.
Guide limitations
- The README gives dataset-specific checkpoint paths; select the path corresponding to the dataset used.
- The instructions do not establish results for other datasets or guarantee successful execution in a given environment.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- wgcban/HyperTransformer · Official homepage title · retained in its source language