LightlyTrain now supports the full object detection model training pipeline: from pretraining on unlabeled data, to supervised fine-tuning, and inference. Built on top of DINOv3, the models achieve excellent detection performance and are easy to fine-tune on any dataset.
.png)
LightlyTrain 0.12.0 introduces four new LTDETR models based on the DINOv3 ConvNeXt backbones, ranging from the tiny size with 60M parameters to large with 230M parameters. The models achieve excellent results on the COCO dataset and the trained model weights are readily available for inference and fine-tuning.
For more information, check out LightlyTrain Object Detection Docs.
Getting started with LTDETR only requires a few lines of code:
import lightly_train
if __name__ == "__main__":
# Train
lightly_train.train_object_detection(
out="out/my_experiment",
model="dinov3/convnext-tiny-ltdetr-coco",
data={
"path": "base_path_to_your_dataset",
"train": "images/train2012",
"val": "images/val2012",
"names": {
0: "person",
1: "bicycle",
# ...
},
}
)
# Load model
model = lightly_train.load_model("dinov3/convnext-tiny-ltdetr-coco")
# Predict
prediction = model.predict("image.jpg")
Ready to try it? Head over to the LightlyTrain docs to start experimenting: 👉Object Detection Docs
Stay tuned for more updates including smaller and faster models!
Self-Supervised Pretraining
Leverage self-supervised learning to pretrain models
Smart Data Capturing on Device
Find only the most valuable data directly on device
.png)
.svg)