Search papers, labs, and topics across Lattice.
Université Paris-Saclay, ENS Paris-Saclay, Centre Borelli, Institut Universitaire de France
7
0
7
Early transformer layers act as in-context PCA engines, compressing high-dimensional visual demonstrations into a universal, low-dimensional subspace where classification becomes trivially separable.
Season-invariant feature matching can drastically cut down geometric errors in satellite imagery, making GCP-free refinement more reliable than ever.
SeasonStereo achieves LiDAR-level accuracy in 3D reconstruction from multi-date satellite images without the need for expensive aligned datasets or ground-truth labels.
Removing stochastic noise from statistical regularizers leads to faster convergence and superior performance in Self-Supervised Learning tasks.
Most state-of-the-art vehicle re-identification methods fail to generalize beyond training data, revealing critical vulnerabilities in their robustness to unseen vehicle types.
Gaussian embeddings might be suboptimal for self-supervised learning: enforcing hyperspherical uniformity in learned representations yields substantial gains in texture retrieval and ImageNet classification.
Stop struggling with scarce industrial datasets: IRIS-v2 offers a comprehensive multimodal resource for automating the alignment of functional schematics with real-world scene data.
