Skip to content
ModelListedExternal
Browse files

Synthetic Usg Afast Vet Classifier

Two-stage transfer learning on koscielnamarta/synthetic-usg-afast-vet.

Sourcehuggingface.co/koscielnamarta/synthetic-usg-afast-vet-classifier/tree/78a160d9b51e691da5c86f297a737818bd1b4706
KindOtherBase modelMacDunno/TinyUSFMVersionRevision 78a160d

Model source

Kind
Other
File size
21.1 MB
Base model
MacDunno/TinyUSFM
Version
vRevision 78a160d
Updated
Sep 28, 2026
Downloads
0
Source
Hugging Face

Source description

TinyUSFM fine-tuned for synthetic veterinary AFAST classification

Two-stage transfer learning on koscielnamarta/synthetic-usg-afast-vet.

StageParamsBest val_accEpochsLR
I1 linear probing1,156 (0.0%)0.723850.001
I2 unfreeze 1 blocks446,020 (8.1%)0.998880.0001

Trained exclusively on synthetic data -- not clinically usable. Source: github.com/koscielnamarta/vet-eye-ai-usg-demo

Sources

1 source
Hugging FacePrimaryVersion 78a160d9b51e691da5c86f297a737818bd1b4706
huggingface.co/koscielnamarta/synthetic-usg-afast-vet-classifier/tree/78a160d9b51e691da5c86f297a737818bd1b4706

Verified Sep 28

Download files

Model artifacts

2 artifacts

checkpoints/best_i1.pt

pt · 21.1 MB · SHA-256 f59a26971a99…211f · Hugging Face

Download file

checkpoints/best_i2.pt

pt · 21.1 MB · SHA-256 65917f71ee93…ffc9 · Hugging Face

Download file

Source excerpts

2 excerpts
ReadmeSep 28huggingface.co/koscielnamarta/synthetic-usg-afast-vet-classifier

--- license: mit tags: - ultrasound - veterinary - AFAST - tinyusfm - image-classification library_name: pytorch pipeline_tag: image-classification datasets: - koscielnamarta/synthetic-usg-afast-vet metrics: - accuracy - f1 base_model: MacDunno/TinyUSFM --- # TinyUSFM fine-tuned for synthetic veterinary AFAST classification Two-stage transfer learning on [koscielnamarta/synthetic-usg-afast-vet](https://huggingface.co/datasets/koscielnamarta/synthetic-usg-afast-vet). | Stage | Params | Best val_acc | Epochs | LR | |---|---|---|---|---| | I1 linear probing | 1,156 (0.0%) | 0.7238 | 5 | 0.001 | | I2 unfreeze 1 blocks | 446,020 (8.1%) | 0.9988 | 8 | 0.0001 | Trained exclusively on synthetic data -- not clinically usable. Source: [github.com/koscielnamarta/vet-eye-ai-usg-demo](https://github.com/koscielnamarta/vet-eye-ai-usg-demo)

NoteSep 28huggingface.co/koscielnamarta/synthetic-usg-afast-vet-classifier

Source context: 0 downloads · 0 likes · Pipeline image-classification · Library pytorch · Repo koscielnamarta/synthetic-usg-afast-vet-classifier