# NVlabs/RADIO

Official repository for "AM-RADIO: Reduce All Domains Into One"

Repository: https://github.com/NVlabs/RADIO
Canonical: https://ross.abutalabs.com/products/radio
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-05-29T07:44:42+00:00

## Health v2 (maintenance only)
Score: 64/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 84, release rhythm 35, longevity 71
- inputs: {"age_days": 999, "days_push": 96, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1933, forks 74 (observed 2026-08-28T04:05:56.372813+00:00)

## What it is
Official PyTorch implementation of AM-RADIO and its successors (RADIOv2.5, C-RADIOv4), agglomerative vision foundation models distilled from multiple teachers like SigLIP2, DINOv3, and SAM. It provides pretrained vision encoders that produce general-purpose image features for downstream computer vision tasks.

## Use cases
- extract general-purpose image features for downstream vision tasks
- use a single vision encoder across segmentation, classification, and depth tasks
- replace CLIP/DINOv2 backbones in my vision pipeline
- fine-tune a distilled vision foundation model
- run multi-teacher vision model distillation research
- get dense visual features for detection and segmentation

## When to choose
- you need one vision backbone that combines strengths of CLIP, DINO, and SAM-style models
- you want strong off-the-shelf dense features for segmentation or detection
- you are researching vision model distillation or foundation model agglomeration

## When to avoid
- you need a commercially permissive license for older RADIO versions (early releases were non-commercial)
- you need text/multimodal encoders rather than a vision-only model
- you lack GPU resources, as these are large transformer models

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing
- domain: computer-vision, deep-learning, machine-learning
- platform: python
- tags: vision-foundation-model, model-distillation, pytorch, research-code, nvidia, feature-extraction, gpu

## Member repositories
- NVlabs/RADIO (main) score 64

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:56.372813+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:08:33.207280+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/RADIO (fetched 2026-08-28T04:05:56.372813+00:00, sha f815cdf73757)
- Data as of 2026-08-30T08:39:29.467469+00:00.