ImageNet-1k linear probe on DINOv3 ViT-L/16 at 512 px
A linear ImageNet-1k classifier on the CLS token of frozen DINOv3 ViT-L/16 at 512 × 512 px, released with CanViT.
CanViT, the Canvas Vision Transformer, is an active-vision foundation model: it sees a scene through a sequence of glimpses and remembers it on a scene-wide canvas.
Paper (NeurIPS 2026) · Code · Project page · All checkpoints
Usage
uv add "dinov3-in1k-probes @ git+https://github.com/m2b3/dinov3-in1k-probes.git" transformers pillow
import torch
from PIL import Image
from transformers import AutoModel
from dinov3_in1k_probes import DINOv3LinearClassificationHead
from dinov3_in1k_probes.data import make_val_transform
dinov3 = AutoModel.from_pretrained("facebook/dinov3-vitl16-pretrain-lvd1689m").eval()
probe = DINOv3LinearClassificationHead.from_pretrained("canvit/dinov3-vitl16-lvd1689m-in1k-512x512-linear-clf-probe").eval()
image = make_val_transform(512)(Image.open("scene.jpg").convert("RGB")).unsqueeze(0) # [1, 3, 512, 512]
with torch.inference_mode():
logits = probe(dinov3(pixel_values=image).last_hidden_state[:, 0]) # [1, 1000], from the CLS token
Results
| ImageNet-1k validation top-1 | 87.44% |
| ImageNet-ReaL validation top-1 | 90.42% |
| ImageNet-1k validation top-5 | 97.60% |
Details
| Features | the CLS token of facebook/dinov3-vitl16-pretrain-lvd1689m, 512 px images |
| Probe | a linear layer, 1024 to 1000 |
| Training | 12,520 steps at batch size 2048 on the ImageNet-1k training set; AdamW, peak learning rate 0.00298, weight decay 0.0311, 1,252-step warmup; every hyperparameter is in config.json |
| Code | m2b3/dinov3-in1k-probes |
License
MIT. The probe reads features of DINOv3 ViT-L/16, which is released under the DINOv3 License.
Citation
@inproceedings{berreby2026canvit,
title={CanViT: Toward Active-Vision Foundation Models},
author={Berreby, Yoha{\"i}-Eliel and Du, Sabrina and Durand, Audrey and Krishna, B. Suresh},
booktitle={Advances in Neural Information Processing Systems},
year={2026},
eprint={2603.22570},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2603.22570}
}