When ML teams send their data to companies like Scale.ai for labeling, most can only afford to label 1% or less of their datasets. But today they don’t have a good way to pick which 1% to label. We help them pick the best 1% of their data to label. By labeling the most representative data, they significantly improve model accuracy at the same cost.
Self-supervised pretraining framework for vision models that requires no labels, supporting distillation, fine-tuning, and integration with models like DINOv3.
Provides AI training data services for LLMs, AI agents, and computer vision.
What customers say
“Lightly enabled us to improve our ML data pipeline in all regards: Selection, Efficiency, and Functionality. This allowed us to cut customer onboarding time by 50% while achieving better model performance.”
CustomerLightly customer
“By integrating Lightly into our existing workflow, we achieved a 90% reduction in dataset size and doubled the efficiency of our deployment process. The tool’s seamless implementation significantly enhanced our data pipeline.”
CustomerLightly customer
“The pretrained models were low in performance. The color scheme is probably the reason, they just don’t transfer well to ash-RGB. This is why we decided to give LightlyTrain distillation a try.”
CustomerLightlyTrain customer
Case study
Rabot used Lightly to select relevant packing-station frames for customer-specific model retraining. The vendor reports a 50% reduction in retraining time, doubled onboarding speed and a 10% improvement in model…
Aigen integrated automated Lightly selection and expert review into its crop-and-weed image pipeline before labeling. The vendor reports doubled deployment efficiency and an 80% dataset-size reduction in the results…
EUROCONTROL used LightlyTrain for reproducible model-distillation workflows on ash-RGB satellite imagery for contrail segmentation. Lightly reports early improvements for YOLO-based models, while noting that formal…
LightlyTrain 0.16.0 adds real-time detection and depth estimation inference
LightlyTrain 0.16.0 introduces LTDETRv2, a suite of real-time transformer-based detectors optimized for edge deployment. The release also adds support for depth estimation inference using Depth Anything V2 and V3 models, and expands SAHI (Slicing Aided Hyper Inference) support to EoMT instance segmentation. Additionally, a new benchmarking command has been integrated to allow users to compare model latency, footprint, and accuracy across different backends, including PyTorch, ONNX Runtime, and TensorRT, facilitating more informed deployment decisions.
