Cohere Labs - Eren Ata, IPB-CeDRI Lab

Gargantua is a physics-aware Video-to-Video (V2V) model designed for counterfactual object removal. Built as a LoRA adapter for the CogVideoX architecture, it utilizes a novel 4-value Quadmask to understand and erase not just targeted objects, but their physical consequences such as cast shadows or collapsing structures. Trained on a custom, high-fidelity Unity/PhysX dataset, Gargantua achieves a measurable reduction in temporal flickering, offering a highly stable video editing pipeline.

Eren ATA, a Big Data Analysis student at Manisa Celal Bayar University and currently a research intern at IPB-CeDRI Lab (Digitalization and Intelligent Robotics Research Centre). My work focuses on LLMOps, evaluating LLMs within VR/Unity environments, and processing large scale datasets.

Add event to calendar

Apple Google Office 365 Outlook Outlook.com Yahoo