Models

Features models

16 features models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.

Input: one RGB image. Output: dense patch features and a pooled image feature.

Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 224x224: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory). – = not measured.

Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.

modelcommercialkernels time (ms),
speed-up vs original		kernels peak memory (MiB, reserved),
ratio to original
dinov2-vitb14yes1.478, 3.0x290, 0.75x
dinov2-vitb14-regyes1.501, 2.9x288, 0.75x
dinov2-vitg14yes13.156, 2.6x2318, 0.53x
dinov2-vitg14-regyes12.98, 2.6x4430, 1.0x
dinov2-vitl14yes4.593, 2.1x714, 0.58x
dinov2-vitl14-regyes4.6, 2.3x714, 0.58x
dinov2-vits14yes0.848, 3.9x148, 1.1x
dinov2-vits14-regyes0.814, 5.0x142, 1.1x
dinov3-convnext-baseyes2.179, 3.4x214, 0.54x
dinov3-convnext-largeyes3.593, 2.7x494, 0.55x
dinov3-convnext-smallyes1.99, 3.0x152, 0.55x
dinov3-convnext-tinyyes1.18, 2.5x112, 0.69x
dinov3-vitb16yes1.404, 5.0x252, 0.65x
dinov3-vith16plusyes9.301, 2.8x1694, 0.52x
dinov3-vitl16yes3.279, 5.4x674, 0.56x
dinov3-vits16yes0.903, 7.2x106, 0.8x

All models

Modes, licences, training data and outputs per model

Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).

modelmodescode licenceweights licencetraining datacommercialoutputs
dinov2-vitb14portable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vitb14-regportable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vitg14portable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vitg14-regportable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vitl14portable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vitl14-regportable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vits14portable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov2-vits14-regportable, kernelsApache-2.0Apache-2.0unstatedyeslast_hidden_state, pooler_output, hidden_states
dinov3-convnext-baseportable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output
dinov3-convnext-largeportable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output
dinov3-convnext-smallportable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output
dinov3-convnext-tinyportable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output
dinov3-vitb16portable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output, hidden_states
dinov3-vith16plusportable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output, hidden_states
dinov3-vitl16portable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output, hidden_states
dinov3-vits16portable, kernelsLicenseRef-DINOv3-LicenseLicenseRef-DINOv3-Licenseunstatedyeslast_hidden_state, pooler_output, hidden_states

Training datasets

iddatasetlicencecommercial use