Michaël Clément

Achieving a three-order speedup in stress field reconstruction while maintaining high fidelity could revolutionize multi-scale simulations in materials science.

Vision-language models can get a 20% segmentation boost by learning to parse *non-semantic* textual descriptions of shape and texture.

Hierarchical superpixels can now be generated with state-of-the-art accuracy and regularity, opening new avenues for multi-scale image analysis.