Merve Kocabas

I am a Master’s student in Machine Learning at the University of Tübingen, working with Prof. Andreas Geiger and Gege Gao in the Autonomous Vision Group. I am currently a visiting student at the Georgia Institute of Technology, where I work with Prof. Sara Fridovich-Keil in the VOILA Lab. My studies are supported by the ELIZA Master Scholarship, and my master’s thesis is supported by a PROMOS scholarship. Before my master’s studies, I completed my undergraduate education at Middle East Technical University, where I studied Computer Engineering and Aerospace Engineering. I am always happy to connect, so feel free to reach out!

Email  /  Github  /  LinkedIn

Recent Work

Abstract: Diffusion and flow-based generative models typically treat generation as an opaque mapping from noise to image. Although recent methods have begun to exploit trajectory structure for improved quality, intermediate states remain uninterpretable and inaccessible to users. We propose Trajectory Forcing (TF), a framework that elevates the generative trajectory from a hidden computational process to an explicit, controllable object. Generation is organized as a sequence of semantically structured stages, progressing from global layout through object-level parts to fine-grained detail, where every intermediate state can be decoded, inspected, and edited. We construct coarse-to-fine teacher hierarchies via unsupervised clustering in pretrained, semantically meaningful representation spaces (e.g., DINOv2), and train a hierarchy-conditioned model using one-step flow matching at each level. We further introduce trajectory-aware evaluation measures that quantify structural consistency and local controllability beyond standard metrics such as FID. Experiments show that TF achieves competitive sample quality while producing structurally coherent, decodable intermediate states that support localized editing at each semantic level. By shifting the modeling focus from the final model outputs to generative dynamics, Trajectory Forcing enables controllable, trajectory-aware image synthesis.

Total … visits since June 2026

Website template by Jon Barron.