Contamination MapVision / Model
DINOv3 (web / LVD-1689M models)
- Organisation
- Meta AI
- Family
- DINO · version 3
- Released
- 2025-08-13 (arXiv v1 submission date (https://arxiv.org/abs/2508.10104v1))
- Checkpoints
- ViT-7B/16, ViT-S/16, ViT-S+/16, ViT-B/16, ViT-L/16, ViT-H+/16, ConvNeXt-T, ConvNeXt-S, ConvNeXt-B, ConvNeXt-L
- Weights
- Open · DINOv3 License
- Links
- PaperModel cardCode
Training data disclosure: Partial list
Level determined for 0% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen0
- ?Unknown8
“We retrieve images from the data pool that are similar to those from selected seed datasets, creating a dataset that covers visual concepts relevant for downstream tasks.”
“For the third part, we use raw publicly available computer vision datasets including ImageNet1k (Deng et al., 2009), ImageNet22k (Russakovsky et al., 2015), and Mapillary Street-level Sequences (Warburg et al., 2020).”
Training corpora
-
LVD-1689M
· 1 source (complete list) · pre-training
Evidence that DINOv3 (web / LVD-1689M models) uses it
“10 models pretrained on web data (LVD-1689M dataset)”
raw.githubusercontent.com · readme · Model Details · retrieved 2026-10-15 Contents of this corpus
-
web images from public posts on Instagram (~17B-image pool, k-means balanced subset) (broad web data: may contain benchmark test sets) · split: unknown
Collection-level entry. Any image posted publicly on Instagram could be in the pool.
Evidence
“We build our large-scale pre-training dataset by leveraging a large data pool of web images collected from public posts on Instagram.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 “This results in a curated subset of 1,689 million images (named LVD-1689M) that guarantees a balanced coverage of all visual concepts appearing on the web.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 “a curated dataset of 1,689 millions of images extracted from a large data pool of 17 billions web images collected from public posts on Instagram”
raw.githubusercontent.com · readme · Training Data · retrieved 2026-10-15
-
web images from public posts on Instagram (~17B-image pool, k-means balanced subset) (broad web data: may contain benchmark test sets) · split: unknown
Other declared training data
-
retrieval-curated images from the Instagram pool (seed datasets not named) (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“We retrieve images from the data pool that are similar to those from selected seed datasets, creating a dataset that covers visual concepts relevant for downstream tasks.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 -
ImageNet1k · split: unknown
Evidence
“For the third part, we use raw publicly available computer vision datasets including ImageNet1k (Deng et al., 2009), ImageNet22k (Russakovsky et al., 2015), and Mapillary Street-level Sequences (Warburg et al., 2020).”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 “In our training, homogeneous batches from ImageNet1k account for 10% of training.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 -
ImageNet22k · split: unknown
Evidence
“For the third part, we use raw publicly available computer vision datasets including ImageNet1k (Deng et al., 2009), ImageNet22k (Russakovsky et al., 2015), and Mapillary Street-level Sequences (Warburg et al., 2020).”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 -
Mapillary Street-level Sequences · split: unknown
Evidence
“For the third part, we use raw publicly available computer vision datasets including ImageNet1k (Deng et al., 2009), ImageNet22k (Russakovsky et al., 2015), and Mapillary Street-level Sequences (Warburg et al., 2020).”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15
Benchmark cells
Industrial anomaly detection
Unknown MVTec AD · training data not disclosed in enough detail
Unknown MVTec AD 2 · training data not disclosed in enough detail
Unknown MVTec LOCO AD · training data not disclosed in enough detail
Unknown VisA · training data not disclosed in enough detail
Unknown Real-IAD · training data not disclosed in enough detail
Unknown MPDD · training data not disclosed in enough detail
Unknown BTAD · training data not disclosed in enough detail
Unknown KolektorSDD2 · training data not disclosed in enough detail
Notes
The 'including' wording may mean the third part has more raw datasets than these three. No benchmark deduplication is described. The two satellite models trained on SAT-493M (Maxar imagery) are excluded from this entry because they use different data. Splits of ImageNet1k and ImageNet22k are not stated.
Sources
- DINOv3 · paper · retrieved 2026-10-15
- DINOv3 MODEL_CARD.md · readme · retrieved 2026-10-15