Fast image generators adapted to improve visual model training accuracy
Fast and Useful: Adapting Distilled Generators for Visual Learning
Computer Vision and Pattern Recognition
Summary
Some fast image-generating AI models produce pictures quickly but these pictures don’t train other AI models well. The authors developed a way to tweak these fast models so they produce images that help train better visual classifiers, without slowing down much. They adjust just one step of the image creation process using info from a slower, more detailed model. This method helps AI models recognize images more accurately than before, even on big datasets like ImageNet.
What this means in practice
- •For computer vision engineers: Train higher-accuracy image classifiers faster using adapted distilled image generators that produce better training images with minimal overhead.
- •For machine learning platform teams: Integrate adapted fast image generators into data pipelines to improve training data quality without increasing computational cost significantly.
Authors
Patryk Bartkowiak, Dominik Michels, Sören Pirk, Wojciech Pałubicki
Abstract
Distilled text-to-image models generate images quickly, but classifiers trained on those images can perform substantially worse than classifiers trained on images from their slower base models. We adapt released distilled generators to produce better training data while retaining fast generation. We fit a low-rank update to the student's first prediction so that it reaches an intermediate state produced by its unguided base, then disable the update and let the unchanged student finish generation. The base supplies training targets and is absent at inference. Guided targets and shorter jumps give classifiers lower accuracy in our ablations, showing why the state used for supervision matters. Across three generator families, the adapted images train better classifiers than images from the released students. With class-name templates at full ImageNet-1k scale, ResNet-50 top-1 rises from 8.4% to 34.2% for FLUX.2-klein and from 12.7% to 32.1% for SD-Turbo, using two and three denoiser evaluations per image, respectively. Fixed adapters also improve results on additional classes, fine-grained datasets, and other learner architectures and training recipes. Template-based adaptation remains useful against richer-prompt alternatives, while the benefit of combining adaptation with LLM generation prompts depends on the generator. Adapting a single generation step makes fast generators substantially more useful for training visual models.