728x90
논문 리뷰 - Self-training with Noisy Student improves ImageNet classification (byeongjokim.github.io)
논문 리뷰 - Self-training with Noisy Student improves ImageNet classification
ImageNet에 속하지 않은 unlabeled images 까지 이용하여 ImageNet 정확성의 SOTA를 갱신하며, robustness 성능 까지 높이는 방법에 대해 설명한다.
byeongjokim.github.io
728x90
'Machine Learning' 카테고리의 다른 글
| Deep Learning on Small Datasets without Pre-Training using Cosine Loss (0) | 2021.02.01 |
|---|---|
| d-SNE: Domain Adaptation using Stochastic Neighborhood Embedding (0) | 2021.02.01 |
| Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection (0) | 2021.02.01 |
| Relation Networks for Object Detection (0) | 2021.02.01 |
| YOLACT: Real-time Instance Segmentation (0) | 2021.02.01 |