Mohamed Eltahir

Black-box output posteriors across 1D image slices expose where VLMs actually look far more faithfully than internal attention—and can be read out directly to fix half of a model's visual reasoning errors.

Generative Late-Interaction Embeddings (GLIE): k<<N vectors per page learned from the normalized centroids to serve as both a lightweight index and a basis for regenerating the page's full embedding set, with the decoder as its main design surface.

In the most challenging crowd scenarios, a point-based counter outperforms traditional detection and segmentation methods, revealing a critical insight for real-world applications in crowd management.