Adapt better

Solve any computer vision task

Our modular model heads attach onto models like drill bits
to address as many use-cases as possible.

4.8

Audi

diabetic retinopathy

branch retinal vein occlusion

tessellation

DOT 0 0 K J E E A A J 3 1 2 2

• Regression
• Multiclass classification
• Multilabel classification
• Object detection
• Quadrilateral detection
• Keypoint detection
• Semantic segmentation
• Instance segmentation
• Scene text recognition
• Depth estimation
• and more...

Enhance your model even further with advanced features like
self-supervised learning, task conversions, multi-task training, and more.

Ship faster

Data in, model out

ML training automation brings focus back onto
dataset curation, product iteration, and business value creation.

Check out our human-readable data annotation formats.

Tune finer

Leverage pre-trained backbones

Pick the right starting point, whether you need to deploy on
constrained embedded systems or scalable cloud infrastructures.

Mobilenet V3 Small : 0.1 GFLOPs, 67.7% Imagenet accuracy
Mobilenet V3 Large : 0.2 GFLOPs, 74% Imagenet accuracy
Convnext Tiny : 4.5 GFLOPs, 82.5% Imagenet accuracy
Efficientnet V2 Small : 8 GFLOPs, 84.2% Imagenet accuracy
Efficientnet V2 Medium : 25 GFLOPs, 85.1% Imagenet accuracy
Efficientnet V2 Large : 56 GFLOPs, 85.8% Imagenet accuracy
• Mobilenet V3
  Small
• Mobilenet V3
  Large
• Convnext
  Tiny
• Efficientnet V2
  Small
• Efficientnet V2
  Medium
• Efficientnet V2
  Large

We rely on open-source models (along with their Imagenet pre-trained weights)
generously provided by Torchvision and Timm.