🧪 Test?View on arXiv
Expanding the Quality-Coverage Frontier with Competitive Flow Matching
Not provided in the abstract
flow matchingimage generationquality-coveragetraining techniques
2611.07055
Builder Relevance
3h ago80%
Abstract
This paper introduces a method to improve the quality-coverage tradeoff in flow matching generators by implementing competition among predictions during training.
Reality Card
Core Claim
The proposed method significantly improves the quality-coverage frontier, reducing the failure distance metric on ImageNet-256 with minimal compute.
Method / Result
Achieved a reduction in FD_{r,6} from 19.22 to 5.91 at the 67% Recall point on ImageNet-256.
Limitations
The paper does not specify the authors or provide detailed reproducibility guidelines.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.