dinov2-small
by facebook
DINOv2 ViT-S is the smallest variant in Meta's DINOv2 series, offering a 21M-parameter self-supervised vision transformer suitable for resource-constrained feature extraction applications.
4,372,990 ↓ · 72 ♡
image feature extraction models
5 models · ranked by HuggingFace downloads
dinov2-base
by facebook
DINOv2 ViT-B is Meta's self-supervised vision transformer trained on 142M curated images using a combination of DINO and iBOT objectives. It produces strong visual features for dense prediction tasks without any labels during pretraining.
2,843,962 ↓ · 195 ♡
vit_small_patch14_dinov2.lvd142m
by timm
A ViT-Small backbone pre-trained with DINOv2 self-supervised learning on the curated LVD-142M dataset. DINOv2 models learn dense visual features without labels, producing representations that transfer well to segmentation, depth estimation, and retrieval tasks. The small patch14 variant offers a balance between spatial resolution and inference speed.
1,135,437 ↓ · 8 ♡
dinov2-large
by facebook
DINOv2 ViT-L is Meta's large-scale self-supervised vision transformer, offering significantly better visual representations than the base variant at 4x the parameter count. It achieves near-supervised performance on linear probing for ImageNet.
768,331 ↓ · 117 ♡
vit_small_patch16_dinov3.lvd1689m
by timm
vit_small_patch16_dinov3.lvd1689m is a Vision Transformer small model trained with the DINOv3 self-supervised learning method on the LVD-1689M large-scale image dataset. It is distributed through the timm library and targets dense image feature extraction without task-specific fine-tuning. The arxiv:2508.10104 reference points to the DINOv3 methodology paper.
412,309 ↓ · 7 ♡