-
CanViT: Toward Active-Vision Foundation Models
Paper • 2603.22570 • Published • 13 -
canvit/canvitb16-add-vpe-pretrain-g128px-s512px-in21k-dv3b16-2026-02-02
Image Feature Extraction • 98.1M • Updated • 690 • 3 -
canvit/canvitb16-add-vpe-finetune-g128px-s512px-in1k-2026-04-06
Image Classification • 95.9M • Updated • 23 • 1 -
canvit/canvitb16-add-vpe-pretrain-g128px-s512px-in1k-dv3b16-2026-06-22
Image Feature Extraction • 98.1M • Updated • 122
CanViT
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
Learned viewing policies for a frozen CanViT. 2026-07-04 qband: 8 seeds; flagship = s2. Code: github.com/m2b3/CanViT-PyTorch-RL
-
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s0
Reinforcement Learning • 5.68M • Updated • 107 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s1
Reinforcement Learning • 5.68M • Updated • 80 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s2
Reinforcement Learning • 5.68M • Updated • 101 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s3
Reinforcement Learning • 5.68M • Updated • 107
JAX / Flax NNX checkpoints for CanViT. https://github.com/yberreby/CanViT-NNX
Linear segmentation probes trained on frozen CanViT canvas features for ADE20K semantic segmentation (150 classes).
-
canvit/probe-ade20k-40k-s512-c8-in21k
Image Segmentation • 160k • Updated • 4 -
canvit/probe-ade20k-40k-s512-c9-in21k
Image Segmentation • 160k • Updated • 8 -
canvit/probe-ade20k-40k-s512-c10-in21k
Image Segmentation • 160k • Updated • 5 -
canvit/probe-ade20k-40k-s512-c12-in21k
Image Segmentation • 160k • Updated • 8
Linear IN1k classification probes for DINOv3 ViT backbones @ 512x512.
-
canvit/dinov3-vits16-lvd1689m-in1k-512x512-linear-clf-probe
385k • Updated • 22 -
canvit/dinov3-vits16plus-lvd1689m-in1k-512x512-linear-clf-probe
385k • Updated • 21 -
canvit/dinov3-vitb16-lvd1689m-in1k-512x512-linear-clf-probe
769k • Updated • 2.13k -
canvit/dinov3-vitl16-lvd1689m-in1k-512x512-linear-clf-probe
1.03M • Updated • 401
Ablation checkpoints for the CanViT pretraining ablation study (appendix).
-
CanViT: Toward Active-Vision Foundation Models
Paper • 2603.22570 • Published • 13 -
canvit/canvitb16-add-vpe-pretrain-g128px-s512px-in21k-dv3b16-2026-02-02
Image Feature Extraction • 98.1M • Updated • 690 • 3 -
canvit/canvitb16-add-vpe-finetune-g128px-s512px-in1k-2026-04-06
Image Classification • 95.9M • Updated • 23 • 1 -
canvit/canvitb16-add-vpe-pretrain-g128px-s512px-in1k-dv3b16-2026-06-22
Image Feature Extraction • 98.1M • Updated • 122
Linear segmentation probes trained on frozen CanViT canvas features for ADE20K semantic segmentation (150 classes).
-
canvit/probe-ade20k-40k-s512-c8-in21k
Image Segmentation • 160k • Updated • 4 -
canvit/probe-ade20k-40k-s512-c9-in21k
Image Segmentation • 160k • Updated • 8 -
canvit/probe-ade20k-40k-s512-c10-in21k
Image Segmentation • 160k • Updated • 5 -
canvit/probe-ade20k-40k-s512-c12-in21k
Image Segmentation • 160k • Updated • 8
Linear IN1k classification probes for DINOv3 ViT backbones @ 512x512.
-
canvit/dinov3-vits16-lvd1689m-in1k-512x512-linear-clf-probe
385k • Updated • 22 -
canvit/dinov3-vits16plus-lvd1689m-in1k-512x512-linear-clf-probe
385k • Updated • 21 -
canvit/dinov3-vitb16-lvd1689m-in1k-512x512-linear-clf-probe
769k • Updated • 2.13k -
canvit/dinov3-vitl16-lvd1689m-in1k-512x512-linear-clf-probe
1.03M • Updated • 401
Learned viewing policies for a frozen CanViT. 2026-07-04 qband: 8 seeds; flagship = s2. Code: github.com/m2b3/CanViT-PyTorch-RL
-
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s0
Reinforcement Learning • 5.68M • Updated • 107 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s1
Reinforcement Learning • 5.68M • Updated • 80 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s2
Reinforcement Learning • 5.68M • Updated • 101 -
canvit/qpolicy-ade20k-c64-t5-qband-2026-07-04-s3
Reinforcement Learning • 5.68M • Updated • 107
Ablation checkpoints for the CanViT pretraining ablation study (appendix).
JAX / Flax NNX checkpoints for CanViT. https://github.com/yberreby/CanViT-NNX