티스토리 뷰
AI
RandAugment: Practical automated data augmentationwith a reduced search space(NeIPS'2020)
아인샴 2024. 4. 29. 12:25https://arxiv.org/pdf/1909.13719
Abstract
- 요즘 augmentation 자동화 전략 좋지만 별도의 search phase로 인해 학습복잡성과 연산 비용이 크게 증가할 수 있다.
- 이 search phase때문에 모델과 데이터셋 크기에 따른 regularization 강도도 조절할 수 없다.
- auto-augmentation의 경우 주로 적은데이터로 작은모델을 학습하거나 그 데이터를 더 큰 모델에 적용했다.
- 제안하는 방식은 search space를 현저하게 줄여 separate proxy task없이 target task학습을 가능하게 했다.
- parameterization 덕분에 정규화 강도를 데이터셋에 맞게 조절할 수 있다.
- randaugment는 다양한 모델과 데이터셋에 맞게 조정가능하며 모든 이전의 automated augemntations 보다 일치하거나 능가하는 성능을 보여줬다.
- 코드 : github.com/tensorflow/tpu/tree/master/models/official/efficientnet

AA | AutoAugment | RA | Random Augmentation |
PBA | Population Based Augmentation |


'AI' 카테고리의 다른 글
Class-Incremental Domain Adaptation(1)ECCV'20 (0) | 2024.05.14 |
---|---|
Large Scale Incremental Learning'CVPR 2019 (0) | 2024.05.03 |
iCaRL: Incremental Classifier and Representation Learning (0) | 2024.04.17 |
Autoaugment: Learning augmentation policies fromdata(CVPR 2018) (1) | 2024.04.14 |
0405 ProCa (0) | 2024.04.06 |
공지사항
최근에 올라온 글
최근에 달린 댓글
- Total
- Today
- Yesterday
링크
TAG
- max-margin_loss
- 전이학습
- alternatives.log
- 첫논문
- 딥페이크탐지
- 1-stage detection
- tf version porting
- TransferLearning
- torchmodel
- batch란
- textual guidance
- /var/log/tallylog
- tf2to1
- /var/log/alternatives.log
- /var/log/faillog
- Arbitrary image stylization
- neural network applications
- 샴네트워크
- yolov3 #tensorflow_convert
- Group Acitivty Recognition
- Source-free unsupervised domain adaptation
- tensorflow convert
- ctfloader
- Deformable Convolutional Networks
- /var/log/fontconfig.log
- Chatgpt4
- StyleCLIP
- DetectoRS
- 도메인 어댑테이션
- Deformable Part Model
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | 6 | 7 |
8 | 9 | 10 | 11 | 12 | 13 | 14 |
15 | 16 | 17 | 18 | 19 | 20 | 21 |
22 | 23 | 24 | 25 | 26 | 27 | 28 |
29 | 30 |
글 보관함