Train NVDINOv2 on images without labels.
What it does
Trains, exports, and runs inference on NVDINOv2 backbones to produce reusable visual features without labeled training data.
When to use it
Use for training, exporting, or running inference on a TAO NVDINOv2 backbone, including self-supervised vision transformer pretraining.
How to use it
Give it images or model checkpoints; it produces training results, exported models, or inference results in output folders.
What you provide
- Your files
Uses
Good to know
- a credential is hard-coded: encryption_key
Latest review
What you need · 5
Requires Docker for the container-based workflow.
Requires NVIDIA Container Toolkit alongside Docker.
The hardware section specifies at least four GPUs, recommends eight, and requires A100 GPUs with at least 40 GB of VRAM per GPU.
Access · 1
This skill
wandb
Write
Tracks training with Weights & Biases