Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Robust and Generalizable Visual Representation Learning via Random Convolutions

About

While successful for various computer vision tasks, deep neural networks have shown to be vulnerable to texture style shifts and small perturbations to which humans are robust. In this work, we show that the robustness of neural networks can be greatly improved through the use of random convolutions as data augmentation. Random convolutions are approximately shape-preserving and may distort local textures. Intuitively, randomized convolutions create an infinite number of new domains with similar global shapes but random local textures. Therefore, we explore using outputs of multi-scale random convolutions as new images or mixing them with the original images during training. When applying a network trained with our approach to unseen domains, our method consistently improves the performance on domain generalization benchmarks and is scalable to ImageNet. In particular, in the challenging scenario of generalizing to the sketch domain in PACS and to ImageNet-Sketch, our method outperforms state-of-art methods by a large margin. More interestingly, our method can benefit downstream tasks by providing a more robust pretrained visual representation.

Zhenlin Xu, Deyi Liu, Junlin Yang, Colin Raffel, Marc Niethammer• 2020

Related benchmarks

TaskDatasetResultRank
Image ClassificationImageNet-Sketch
Top-1 Accuracy18.1
360
Image ClassificationPACS (test)
Average Accuracy67.5
254
Domain GeneralizationPACS (test)
Average Accuracy86.63
225
Semantic segmentationMapillary (val)
mIoU32.43
153
Cardiac SegmentationACDC (test)
Avg Dice87.94
141
Image ClassificationCIFAR-10-C
Accuracy71.23
127
Semantic segmentationBDD-100K (val)
mIoU30.92
102
Image ClassificationPACS
Accuracy62.76
100
Semantic segmentationSYNTHIA (val)
mIoU24.45
71
Cardiac SegmentationACDC-C (unseen test)
Dice Score79.67
36
Showing 10 of 33 rows

Other info

Follow for update