dinov2-large
by facebook
Self-supervised vision foundation model producing all-purpose visual features
facebook/dinov2-largemixpeek://image_extractor@v1/facebook_dinov2_large_v1Overview
DINOv2 is a self-supervised vision foundation model from Meta AI that learns robust visual features without any labels. Trained on a curated dataset of 142M images (LVD-142M) using a combination of DINO and iBOT objectives, it produces dense features that work across image distributions and tasks without fine-tuning.
On Mixpeek, DINOv2 provides high-quality visual embeddings for similarity search, classification, and dense prediction tasks. Its features are especially strong for fine-grained visual understanding.
Architecture
Vision Transformer (ViT-L/14) with 24 layers, 1024-dim hidden size, 16 attention heads. Trained via self-distillation from a 1B-parameter ViT-g teacher. Includes register tokens to fix attention artifacts in feature maps.
Mixpeek SDK Integration
// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so dinov2-large runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
"https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
{
method: "POST",
headers: {
Authorization: "Bearer API_KEY",
"Content-Type": "application/json",
},
body: JSON.stringify({
collection_id: "col_your_collection",
documents: [
{
document_id: "asset-00412",
// The vector name has to match a vector index on the collection.
vectors: { "image-embedding": yourVector },
payload: { source_key: "archive/2026/asset-00412" },
},
],
}),
},
);
// Managed alternative, if this exact model is not the requirement:
// image_extractor@v1 runs google/siglip-base-patch16-224
// (768-d) over a bucket, with no inference of your own.Capabilities
- Self-supervised visual features without any labels
- 1024-dimensional dense embeddings per patch
- Linear-probe classification at 87.1% ImageNet accuracy (ViT-g)
- Strong on depth estimation, segmentation, retrieval
- Register tokens for clean dense feature maps
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| ImageNet (linear probe) | Top-1 Accuracy | 81.6% | Oquab et al., 2024: Table 1 |
| ADE20k (linear seg.) | mIoU | 49.0 | Oquab et al., 2024: Table 3 |
Performance
Common Pipeline Companions
Explore on Mixpeek
Compare alternatives in this category
Hand-picked tools & platforms compared
Deep-dive technical guide
See how Mixpeek runs models as extractors
Store & search embeddings at scale
Usage-based pricing for pipelines
Compare models, APIs & infrastructure
Specification
Research Paper
DINOv2: Learning Robust Visual Features without Supervision
arxiv.orgBuild a pipeline with dinov2-large
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Run it on your own data, free