Let's Get To Know "CUT" Which Can Create More Beautiful And Clearer Images!

Three main points

Show the usefulness of Contrastive learning in ✓ unsupervised image transformation

✓Contrastive learning Negative samples are sampled from within the input image to maximize the amount of mutual information between patches.

✓Improvement of mAP and FID and further reduction of learning time compared to the conventional methods CycleGAN and GCGAN were demonstrated.

Koshiro Nagano

I am researching how to restore image quality in deep generative models. I also want to be involved in the integration of edge AI in future robot development, and contribute to model weight reduction and SIM2REAL. I am working hard every day to study abroad in AI research such as AI/ML Residency Program in a few years.