multimodal-foundation-models

Tag

Cards List
#multimodal-foundation-models

GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration

Hugging Face Daily Papers · 2026-05-29 Cached

This paper introduces GGT-100K, a dataset of 103,707 image pairs for real-world image restoration, generated by using multimodal foundation models like Nano-Banana-2 to produce high-quality targets from low-quality inputs. Experiments show the dataset improves the generalization of various image restoration models.

0 favorites 0 likes
#multimodal-foundation-models

WildTableBench: Benchmarking Multimodal Foundation Models on Table Understanding In the Wild

Hugging Face Daily Papers · 2026-05-01 Cached

WildTableBench introduces the first question-answering benchmark for real-world table images, revealing that existing multimodal foundation models struggle significantly with structural perception and numerical reasoning, with only one model exceeding 50% accuracy.

0 favorites 0 likes
← Back to home

Submit Feedback