ClassMix: Segmentation-Based Data Augmentation for Semi-Supervised Learning

15 Jul 2020  ·  Viktor Olsson, Wilhelm Tranheden, Juliano Pinto, Lennart Svensson ·

The state of the art in semantic segmentation is steadily increasing in performance, resulting in more precise and reliable segmentations in many different applications. However, progress is limited by the cost of generating labels for training, which sometimes requires hours of manual labor for a single image... Because of this, semi-supervised methods have been applied to this task, with varying degrees of success. A key challenge is that common augmentations used in semi-supervised classification are less effective for semantic segmentation. We propose a novel data augmentation mechanism called ClassMix, which generates augmentations by mixing unlabelled samples, by leveraging on the network's predictions for respecting object boundaries. We evaluate this augmentation technique on two common semi-supervised semantic segmentation benchmarks, showing that it attains state-of-the-art results. Lastly, we also provide extensive ablation studies comparing different design decisions and training regimes. read more

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Semi-Supervised Semantic Segmentation Cityscapes 100 samples labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 54.07% # 8
Semi-Supervised Semantic Segmentation Cityscapes 12.5% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 61.35% # 9
Semi-Supervised Semantic Segmentation Cityscapes 25% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 63.63% # 9
Semi-Supervised Semantic Segmentation Cityscapes 2% labeled ClassMix (DeepLabv2 with ResNet101, MSCOCO pre-trained) Validation mIoU 52.14% # 2
Semi-Supervised Semantic Segmentation Cityscapes 50% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 66.29% # 5
Semi-Supervised Semantic Segmentation Cityscapes 5% labeled ClassMix (DeepLabv2 with ResNet101, MSCOCO pre-trained) Validation mIoU 58.77% # 2
Semi-Supervised Semantic Segmentation Pascal VOC 2012 12.5% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 71.00% # 9
Semi-Supervised Semantic Segmentation Pascal VOC 2012 1% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 54.18% # 5
Semi-Supervised Semantic Segmentation PASCAL VOC 2012 25% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 72.45 # 4
Semi-Supervised Semantic Segmentation Pascal VOC 2012 2% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 66.15% # 7
Semi-Supervised Semantic Segmentation Pascal VOC 2012 5% labeled ClassMix (DeepLab v2 MSCOCO pretrained) Validation mIoU 67.77% # 7

Methods


No methods listed for this paper. Add relevant methods here