ramanakshay/amd-expert-corpus

AMD Expert Corpus

Dataset Description

Dataset Sources

• Repository: git@github.com:AI4VSLab/vlm-finetuning.git
• Paper: Co-Annotator: Expert-Distilled ViT and VLM for Visual and Documentation Guidance in Age-Related Macular Degeneration (UIST '26) [1]
• Related models: ramanakshay/vlm-oct-biomarkers [1]

Dataset Structure

amd-expert-corpus/
├── Images/                    # Macular OCT B-scans (five-line raster; five slices per eye)
├── Heatmap/                   # Expert fixation-density heatmaps (32x32 ViT patch grid)
├── GroupedBiomarker.tfrecord  # Multi-label biomarker annotations, grouped per eye/bundle
└── Groups.pdf                 # Documentation of the group (bundle) structure

• Images/ — Macular OCT B-scans from 231 eyes (104 normal, 127 wAMD), five B-scans per eye [1].
• Heatmap/ — Per-image expert attention targets: fixations detected via velocity-thresholding were converted to fixation-density heatmaps by kernel density estimation, normalized to unit mass and downsampled to the 32×32 ViT patch grid [1].
• GroupedBiomarker.tfrecord — Multi-label biomarker annotations extracted from expert dictations using a prompted extraction model constrained to an ontology derived from OCT5k, reviewed by a retina specialist [1]. The vocabulary contains 12 predefined terms: Drusen, Photoreceptor Degeneration, Pigment Epithelial Detachment, Geographic Atrophy, Choroidal Fold, Epiretinal Membrane, Hyperfluorescent Spots, Intraretinal Fluid, Posterior Vitreous Detachment, Fluid, Subretinal Fluid, Choroidal Neovascularization [1].
• Groups.pdf — Documentation of how scans are grouped into per-eye bundles.

Citation

BibTeX:

@inproceedings{li2026coannotator,
  author    = {Li, Ziheng 'Leo' and Freeman, Benjamin and Raman, Akshay and Rajkumar, Kavin Aravindhan and Fang, Xinxin and Srivastava, Rishabh and Feiner, Steven and Thakoor, Kaveri A.},
  title     = {Co-Annotator: Expert-Distilled ViT and VLM for Visual and Documentation Guidance in Age-Related Macular Degeneration},
  booktitle = {Proceedings of the 39th Annual ACM Symposium on User Interface Software and Technology (UIST '26)},
  year      = {2026},
  publisher = {ACM},
  address   = {New York, NY, USA}
}

APA:

Li, Z. L., Freeman, B., Raman, A., Rajkumar, K. A., Fang, X., Srivastava, R., Feiner, S., & Thakoor, K. A. (2026). Co-Annotator: Expert-Distilled ViT and VLM for Visual and Documentation Guidance in Age-Related Macular Degeneration. In Proceedings of the 39th Annual ACM Symposium on User Interface Software and Technology (UIST '26). ACM.

Dataset Card Contact

Ziheng 'Leo' Li — zihengleoli@cs.columbia.edu