本文共 2523 字,大约阅读时间需要 8 分钟。
目录
Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation (arxiv 2020.12)
idea:
- perform a systematic study of the Copy-Paste augmentation (e.g., [13, 12]) for instance segmentation where we randomly paste objects onto an image
- we show Copy-Paste is additive with semi-supervised methods that leverage extra data through pseudo labeling (e.g. self-training).
Introduction:
- a simple strategy of randomly picking objects and pasting them at random locations on the target image provides a signifificant boost on top of baselines across multiple settings.
- we show that the Copy-Paste augmentation results in better features for the two-stage training procedure typically used in the LVIS benchmark (long-tail dataset)
Learning Invariances in Neural Networks (nips2020)
Introduction:
- The ability to learn constraints or symmetries is a foundational property of intelligent systems
- we provide an extremely simple and practical approach to automatically discovering invariances and equivariances
- Our approach operates by learning a distribution over augmentations, then training with augmented data, leading to the name Augerino.
- Augerino (1) can learn both invariances and equivariances over a wide range of symmetry groups, including translations, rotations, scalings, and shears;
- (2) can discover partial symmetries, such as rotations not spanning the full range from [-π, π];
- (3) can be combined with any standard architectures, loss functions, or optimization algorithm with little overhead;
- (4) performs well on regression, classifification, and segmentation tasks, for both image and molecular data
Method:
- A simple way of constructing a model invariant to a given group of transformations is to average the outputs of an arbitrary model for the inputs transformed with all the transformations in the group
- Augerino discovers invariances by learning θ from training data alone
Self-supervised Pre-training with Hard Examples Improves Visual Representations (arxiv2020)
idea:
- present a modeling framework that unifies existing SSP methods as learning to predict pseudo-labels.
- we propose new data augmentation methods of generating training examples whose pseudo-labels are harder to predict than those generated via random image transformations
- use adversarial training and CutMix to create hard examples (HEXA) to be used as augmented views for MoCo-v2 and DeepCluster-v2
-
hard examples are instrumental in improving the generalization of the pre-trained models. -
如何根据自监督任务去设计更好的data augmentation方法
转载地址:http://cfeqf.baihongyu.com/