Models by task

Every family the library ships, generated from the v1.6.0 model registry: 111 families across 20 tasks.

Detection

46 families

Boxes around objects. The task most of the library is built for.

⭐RF-DETR

n, s, m, l for detection and oriented boxes; n through xx for segmentation; x for pose

docs weights

⭐YOLOv9

yolo9: t, s, m, c at 640 px; yolo9_p2: t, s at 640 px

docs weights

D-FINE

n, s, m, l, x at 640 px

docs weights

DEIM

deim: n, s, m, l, x at 640 px

docs weights

EdgeCrafter

s, m, l, x at 640 px

docs weights

GTR

docs weights

RT-DETR

rtdetr: r18, r34, r50, r50m, r101, l, x at 640 px; rtdetrv2: r18, r34, r50, r50m, r101 at 640 px; rtdetrv4: s, m, l, x at 640 px

docs weights

TinyFormer

docs weights

YOLO-NAS

s, m, l at 640 px

docs weights

Dome-DETR

s, m, l at 800 px

docs weights

PicoDet

s, m, l at 320 to 640 px

docs weights

PP-YOLOE

s, m, l, x at 640 px

docs

RTMDet

t, s, m, l, x at 640 px

docs weights

YOLOv7

b at 640 px

docs weights

YOLOX

n, t, s, m, l, x at 416 to 640 px

docs weights

Florence-2

base, large at 768 px

docs weights

Gemma 4

e2b, e4b at 896 px

docs weights

Grounding DINO

t, b at 800 px

docs weights

InternVL3

1b, 2b, 8b at 448 px

docs weights

Kosmos-2

224 at 224 px

docs weights

LFM2-VL

450m, 3b at 512 px

docs weights

LibreMODUS

docs weights

LocateAnything

3b at 2500 px

docs weights

Moondream

2, 3 at 384 to 512 px

docs weights

North Micro Vision

docs

OMDet-Turbo

t at 640 px

docs weights

OV-DEIM

s, m, l at 640 px

docs weights

OWLv2

b16, l14 at 960 to 1008 px

docs weights

Qwen3-VL

2b, 4b, 8b at 1024 px

docs weights

SenseNova-Vision

7b at 1024 px

docs weights

SmolVLM2

500m at 512 px

docs weights

CenterNet

resdcn18, dla34 at 512 px

docs weights

Deformable DETR

r50ss, r50ssdc5, r50, r50refine, r50twostage at 800 px

docs weights

DETR

r50, r50dc5, r101, r101dc5 at 800 px

docs weights

DINO-DETR

r50, r50s5, swinl at 800 px

docs weights

EfficientDet

docs weights

Faster R-CNN

n, s, m, l at 320 to 800 px

docs weights

FCOS

r50 at 800 px

docs weights

LW-DETR

t, s, m, l, x at 640 px

docs weights

Mask R-CNN

r50 at 800 px

docs weights

RetinaNet

r50, r50v2 at 800 px

docs weights

SSD

300 at 300 px

docs weights

YOLOv1

t, b at 448 px

docs weights

YOLOv2

t, b at 416 to 608 px

docs weights

YOLOv3

t, b, spp at 416 to 608 px

docs weights

YOLOv4

t, b at 416 to 608 px

docs weights

Segmentation

14 families

Per-instance masks rather than boxes.

⭐RF-DETR

n, s, m, l for detection and oriented boxes; n through xx for segmentation; x for pose

docs weights

D-FINE

n, s, m, l, x at 640 px

docs weights

EdgeCrafter

s, m, l, x at 640 px

docs weights

GTR

docs weights

RTMDet

t, s, m, l, x at 640 px

docs weights

EdgeTAM

docs weights

MobileSAM

tiny at 1024 px

docs weights

PicoSAM3

pico at 96 px

docs weights

SAM

base, large, huge at 1024 px

docs weights

SAM 2

docs weights

SAM 3

large at 1008 px

docs weights

SenseNova-Vision

7b at 1024 px

docs weights

EoMT

s, b, l at 512 px

docs weights

Mask R-CNN

r50 at 800 px

docs weights

Keypoints

7 families

Skeletons and landmarks on each detected instance.

⭐RF-DETR

n, s, m, l for detection and oriented boxes; n through xx for segmentation; x for pose

docs weights

EdgeCrafter

s, m, l, x at 640 px

docs weights

GTR

docs weights

YOLO-NAS

s, m, l at 640 px

docs weights

SenseNova-Vision

7b at 1024 px

docs weights

DEKR

w32 at 640 px

docs

HRNet

docs weights

Classification

15 families

One label for the whole image, and the backbones behind everything else.

ConvNeXt

t, s, b at 224 px

docs weights

ConvNeXt V2

docs weights

DINOv2

n, s, m, l at 518 px

docs weights

EfficientNetV2

b0, b1, b2, b3 at 224 to 300 px

docs weights

MobileNetV4

s, m, l at 224 to 256 px

docs weights

ResNet

18, 34, 50, 101 at 224 px

docs weights

V-JEPA 2

l256, h256, g256, g384 at 256 to 384 px

CLIP

Perception Encoder

SigLIP2

AlexNet

b at 224 px

docs weights

Swin Transformer

t, s, b, l at 224 px

docs weights

VGG

16, 19, 16bn, 19bn at 224 px

docs weights

ViT

ti, s, b, l at 224 px

docs weights

DeiT

t, s, b at 224 px

docs weights

Oriented boxes

4 families

Rotated boxes, for aerial imagery and anything that does not sit axis-aligned.

⭐RF-DETR

n, s, m, l for detection and oriented boxes; n through xx for segmentation; x for pose

docs weights

GTR

docs weights

RT-DETR

rtdetr: r18, r34, r50, r50m, r101, l, x at 640 px; rtdetrv2: r18, r34, r50, r50m, r101 at 640 px; rtdetrv4: s, m, l, x at 640 px

docs weights

YOLO-NAS

s, m, l at 640 px

docs weights

Depth

8 families

Distance per pixel from a single image.

GTR

docs weights

LibreMODUS

docs weights

SenseNova-Vision

7b at 1024 px

docs weights

Depth Anything 3

l at 504 px

docs weights

Depth Anything V2

s, b, l, g at 518 px

docs weights

Marigold V2

b at 1024 px

docs weights

MiDaS

s, l at 256 to 384 px

docs weights

ZipDepth

b, bnpu at 384 px

docs weights

Semantic segmentation

10 families

A class for every pixel, without separating instances.

GTR

docs weights

DINOv2

n, s, m, l at 518 px

LingBot-Vision

PP-LiteSeg

SegFormer

U-Net

DeepLabv3

EoMT

s, b, l at 512 px

docs weights

FCN

r50, r101 at 520 px

docs weights

PIDNet

s, m, l at 1024 px

docs weights

Panoptic segmentation

2 families

Semantic and instance masks in one pass.

SenseNova-Vision

7b at 1024 px

docs weights

EoMT

s, b, l at 512 px

docs weights

Points and counting

8 families

Centroids instead of boxes, cheap enough for microcontrollers.

FOMO

docs weights

Florence-2 grounding

base, large at 768 px

docs weights

LocateAnything

3b at 2500 px

docs weights

Molmo2

4b, 8b at 378 px

docs weights

Moondream

2, 3 at 384 to 512 px

docs weights

Qwen3-VL grounding

2b, 4b, 8b at 1024 px

docs weights

SenseNova-Vision

7b at 1024 px

docs weights

ShowUI

2b at 1344 px

docs

Text recognition

2 families

Finding and reading text in an image.

SenseNova-Vision

7b at 1024 px

docs weights

PP-OCRv5

t, l at 960 px

docs weights

Embeddings

7 families

Vectors for search, clustering and retrieval.

DINOv2

n, s, m, l at 518 px

docs weights

V-JEPA 2

l256, h256, g256, g384 at 256 to 384 px

CLIP

Perception Encoder

SigLIP2

LeVJEPA

l at 224 px

docs weights

LibreFaceRec

docs weights

Gaze

1 families

Where a person is looking.

L2CS-Net

r18, r34, r50, r101, r152 at 448 px

docs weights

Edge detection

3 families

Contours and boundaries.

LibreMODUS

docs weights

DexiNed

b at 352 px

docs weights

TEED

t at 352 px

docs weights

Surface normals

3 families

Which way each surface faces.

LibreMODUS

docs weights

Marigold V2

b at 1024 px

docs weights

MoGe-2

s, b, l at 518 px

docs weights

Matting

4 families

Alpha cutouts, hair and edges included.

BEN2

b at 1024 px

docs weights

BiRefNet

t, l at 1024 px

docs weights

FeyNobg

l at 1024 px

docs weights

ViTMatte

s at 512 px

docs weights

Restoration

7 families

Denoising, deblurring and upscaling.

NAFNet

s, l at 256 px

docs weights

DDColor

docs weights

HVI-CIDNet

t at 256 px

docs weights

LaMa

docs weights

QuickSRNet

m2 at 64 px

docs weights

Real-ESRGAN

x4, x2, x4t at 64 px

docs weights

SwinIR

s, m, l at 64 px

docs weights

Mesh

1 families

3D geometry from an image.

SAM 3D Body

d3, h at 512 px

docs weights

Robot policies

1 families

Action chunks from camera observations and robot state.

LibreVLA

smolvla: base at 512 px; act_policy: base at 224 px; diffusion_policy: base at 224 px

docs

Albedo

1 families

Intrinsic surface color in linear RGB.

Marigold V2

b at 1024 px

docs weights

3D detection

4 families

Camera-frame cuboids from a single image.

3D-MOOD

docs weights

DetAny3D

h at 896 px

docs weights

FCOS3D

r101 at 1600 px

docs weights

WildDet3D

l at 1008 px

docs weights

All weights ↗Docs →