Camile Lendering

BenthicFlow generates coherent, large-scale 3D underwater environments in a unified framework, eliminating the need for separate scene stitching.

You can get surprisingly accurate defect segmentation from noisy SAM pseudo-masks by treating SAM as a flawed teacher and correcting its mistakes during training.

Forget finetuning: this training-free anomaly detection method using DINOv2 and PCA achieves state-of-the-art results in few-shot settings.