SuperMix: Supervising the Mixing Data Augmentation

Dabouei, Ali; Soleymani, Sobhan; Taherkhani, Fariborz; Nasrabadi, Nasser M.

doi:10.1109/cvpr46437.2021.01358

Cited by 60 publications

(26 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In other works, authors explore using nonlinear or optimizable interpolation mixup policies, such as PuzzleMix [16], Co-Mixup [15], AutoMix [27], and SAMix [23]. Moreover, mixup methods extend to more than two elements [15,6], and are utilized in contrastive learning to learn discriminative visual representation [14,20,35,23].…”

Section: Related Workmentioning

confidence: 99%

Decoupled Mixup for Data-efficient Learning

Liu¹,

Liu²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Mixup is an efficient data augmentation approach that improves the generalization of neural networks by smoothing the decision boundary with mixed data. Recently, dynamic mixup methods improve previous static policies (e.g., linear interpolation) by maximizing discriminative regions or maintaining the salient objects in mixed samples. We notice that The mixed samples from dynamic policies are more separable than the static ones while preventing models from overfitting. Inspired by this finding, we first argue that there exists an over-smoothing issue in the mixup objective, which focuses on regression the mixing ratio instead of identifying discriminative features. We are therefore prompted to propose a decoupled mixup (DM) loss that can adaptively mine discriminative features without losing smoothness. DM enables static mixup methods to achieve comparable performance with dynamic methods while avoiding heavy computational overhead. This also leads to an interesting objective design problem for mixup training that we need to focus not only on smoothing the decision boundaries but also on identifying discriminative features. Extensive experiments on supervised and semi-supervised learning benchmarks across seven classification datasets validate the effectiveness of DM by equipping with various mixup methods.

show abstract

Section: Related Workmentioning

confidence: 99%

Decoupled Mixup for Data-efficient Learning

Liu¹,

Liu²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…To address both problems, instead of storing whole images from the previous tasks {1, ..., t − 1}, we propose to store an informative portion that we will mix with the images of the current task t. Image mixing is popular for classification [93], [94], [95], [96], [97], [98] yet, to the best of our knowledge, sees limited use for semantic segmentation [99], [100], [101], [102], [103], and has never been considered to design memory-efficient rehearsal learning systems. Formally, given an image I and the corresponding ground truth segmentation maps S t , we define a binary mask Π c such that ∀c ∈ C t : with O c the selected object for class c. By nature, this patch is extremely sparse and can be efficiently stored on disk by modern compression algorithms [104].…”

Section: Object Rehearsalmentioning

confidence: 99%

Tackling Catastrophic Forgetting and Background Shift in Continual Semantic Segmentation

Douillard,

Chen,

Dapogny

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep learning approaches are nowadays ubiquitously used to tackle computer vision tasks such as semantic segmentation, requiring large datasets and substantial computational power. Continual learning for semantic segmentation (CSS) is an emerging trend that consists in updating an old model by sequentially adding new classes. However, continual learning methods are usually prone to catastrophic forgetting. This issue is further aggravated in CSS where, at each step, old classes from previous iterations are collapsed into the background. In this paper, we propose Local POD, a multi-scale pooling distillation scheme that preserves long-and short-range spatial relationships at feature level. Furthermore, we design an entropy-based pseudo-labelling of the background w.r.t. classes predicted by the old model to deal with background shift and avoid catastrophic forgetting of the old classes. Finally, we introduce a novel rehearsal method that is particularly suited for segmentation. Our approach, called PLOP, significantly outperforms state-of-the-art methods in existing CSS scenarios, as well as in newly proposed challenging benchmarks.

show abstract

“…Other works have added random noise to labels to alleviate overfitting [46], some techniques rotate all the images in a class and consider the newly rotated class as distinct from its parent class. Recent works [12,51,41,15,45] are recording better performance values when augmentation strategies are injected within the meta-learning pipeline.…”

Section: Related Workmentioning

confidence: 99%

“…This section builds on the findings of section 5.1, where we established three core data augmentation cases; support, query and task data augmentation. Similar to [41,17,12], we used the CutMix, SelfMix, MixUp, Random Crop and Horizontal Flip augmentation methods on the support, query and task datasets, respectively. We identified the best augmentation combinations that suit a few-shot learner and with our findings, we picked the best strategy to determine which mode of augmentation suits a DBT-regularized fewshot learner.…”

Section: Augmentation Modesmentioning

confidence: 99%

Ortho-Shot: Low Displacement Rank Regularization with Data Augmentation for Few-Shot Learning

Osahor¹,

Nasrabadi²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

In few-shot classification, the primary goal is to learn representations from a few samples that generalize well for novel classes. In this paper, we propose an efficient low displacement rank (LDR) regularization strategy termed Ortho-Shot; a technique that imposes orthogonal regularization on the convolutional layers of a few-shot classifier, which is based on the doubly-block toeplitz (DBT) matrix structure. The regularized convolutional layers of the few-shot classifier enhances model generalization and intra-class feature embeddings that are crucial for few-shot learning. Overfitting is a typical issue for few-shot models, the lack of data diversity inhibits proper model inference which weakens the classification accuracy of few-shot learners to novel classes. In this regard, we broke down the pipeline of the few-shot classifier and established that the support, query and task data augmentation collectively alleviates overfitting in networks. With compelling results, we demonstrated that combining a DBT-based low-rank orthogonal regularizer with data augmentation strategies, significantly boosts the performance of a few-shot classifier. We perform our experiments on the miniImagenet, CIFAR-FS and Stanford datasets with performance values of about 5% when compared to state-of-the-art.

show abstract

SuperMix: Supervising the Mixing Data Augmentation

Cited by 60 publications

References 16 publications

Decoupled Mixup for Data-efficient Learning

Decoupled Mixup for Data-efficient Learning

Tackling Catastrophic Forgetting and Background Shift in Continual Semantic Segmentation

Ortho-Shot: Low Displacement Rank Regularization with Data Augmentation for Few-Shot Learning

Contact Info

Product

Resources

About