Reformulates traditional vision tasks like classification and object detection as a continuous transport process using Discriminative Flow Matching.
March 17, 2026
Original Paper
Discriminative Flow Matching Via Local Generative Predictors
arXiv · 2603.13928
The Takeaway
By moving away from static one-step neural projections toward iterative refinement, this method bridges the gap between generative and discriminative modeling. It allows for robust, iterative inference and the ability to update model blocks either sequentially or in parallel, offering new trade-offs for compute-constrained hardware.
From the abstract
Traditional discriminative computer vision relies predominantly on static projections, mapping input features to outputs in a single computational step. Although efficient, this paradigm lacks the iterative refinement and robustness inherent in biological vision and modern generative modelling. In this paper, we propose Discriminative Flow Matching, a framework that reformulates classification and object detection as a conditional transport process. By learning a vector field that continuously t