Citing AstroVision
AstroVision is an active research and documentation project on machine learning for galaxy morphology classification, multimodal late-fusion models, uncertainty estimation, and physically interpretable computer vision.
AstroVision is not yet a peer-reviewed publication. Until a DOI, Zenodo archive, or manuscript identifier exists, use the provisional citation below for internal references, presentations, and project documentation.
Provisional citation
Use this citation when referring to the project as a research software and documentation effort:
@misc{baker_astrovision_2026,
author = {Baker, Alex},
title = {AstroVision: Multimodal Deep Learning for Galaxy Morphology Classification},
year = {2026},
howpublished = {\url{https://phd-brown.github.io/research-docs/docs/astrovision/}},
note = {Research documentation and active manuscript preparation}
}
Plain-text format
Baker, A. (2026). AstroVision: Multimodal Deep Learning for Galaxy Morphology Classification. Research documentation and active manuscript preparation.
What to cite for each use case
| Use case | Recommended citation |
|---|---|
| General project overview | Provisional AstroVision citation |
| Late-fusion architecture | Provisional citation + link to the Late-Fusion page |
| Benchmark score or model comparison | Provisional citation + Benchmarking / Evaluation Protocol pages |
| Grad-CAM interpretation | Provisional citation + Grad-CAM page |
| Segmentation or morphometry pipeline | Provisional citation + Morphometry / Segmentation pages |
| Future manuscript | Replace provisional citation with the final paper citation once available |
Project components
AstroVision currently includes:
- supervised galaxy morphology classification,
- baseline CNN and EfficientNet models,
- DINOv2 probing and fine-tuning experiments,
- a late-fusion multimodal architecture,
- uncertainty estimation with MC Dropout and conformal prediction,
- Grad-CAM interpretability,
- segmentation-assisted morphometry,
- a bridge to the AstroSpectro scientific documentation ecosystem.
Preferred wording
When referencing the current state of the project, use wording like:
AstroVision is an active research project investigating multimodal deep learning for galaxy morphology classification, with emphasis on uncertainty estimation, interpretability, and physically meaningful evaluation.
Avoid wording that implies peer-reviewed publication until the manuscript has been accepted.
Citation caveats
Do not cite the late-fusion balanced accuracy score (86.3%) without specifying the dataset, split, model version, and evaluation protocol. The score is meaningful only in the context of the documented benchmark setup.
The strongest citation-ready claims should eventually be registered in the transversal Scientific Claims Register.
Future DOI / archive
When the project reaches a stable release, the preferred citation should be updated to include:
| Item | Status |
|---|---|
| GitHub release tag | To add |
| Zenodo DOI | To add |
| Manuscript DOI / arXiv ID | To add |
| Final title | To confirm |
| Final author list | To confirm |
| Versioned documentation URL | To add |
Related pages
| Page | Why it matters |
|---|---|
| Key Results | Main quantitative and scientific outcomes |
| Evaluation Protocol | How reported metrics should be interpreted |
| Benchmarking | Model comparison details |
| Late-Fusion Multimodal | Primary architecture |
| Roadmap | Planned manuscript and release milestones |