Seldinger VideoMAE Lap Chole Broad128

Self-supervised VideoMAE pretraining checkpoint adapted to a mirrored laparoscopic cholecystectomy video corpus.

Training Lineage

Base checkpoint: MCG-NJU/videomae-base
Strict adaptation run: 30 train videos, 200 steps
Broad continuation run: 128 train videos, 500 steps
Final broad metrics: loss_last 0.27357321977615356, loss_mean 0.4516991448639892
Full checkpoint archive with intermediate checkpoints: gs://seldinger-datasets-raw/_runner/surgery-video-training/surgery-videomae-pretrained-broad128-20260714T173724Z/model.tgz

Files

config.json and model.safetensors: final VideoMAEForPreTraining checkpoint
train_metrics.json: final training metrics
training_manifest.jsonl: 128-row broad training manifest used for the continuation run

Intended Use

This is a self-supervised video representation checkpoint for research and downstream surgical-video experimentation. It is not a clinical decision model and should not be used for patient care without downstream validation.

Downloads last month: 10

Safetensors

Model size

94.2M params

Tensor type

F32

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for txmedai/seldinger-videomae-lap-chole-broad128

Base model

MCG-NJU/videomae-base

Finetuned

(703)

this model