Hypercorrelation Squeeze for Few-Shot Segmentation

4 Apr 2021  ·  Juhong Min, Dahyun Kang, Minsu Cho ·

Few-shot semantic segmentation aims at learning to segment a target object from a query image using only a few annotated support images of the target class. This challenging task requires to understand diverse levels of visual cues and analyze fine-grained correspondence relations between the query and the support images. To address the problem, we propose Hypercorrelation Squeeze Networks (HSNet) that leverages multi-level feature correlation and efficient 4D convolutions. It extracts diverse features from different levels of intermediate convolutional layers and constructs a collection of 4D correlation tensors, i.e., hypercorrelations. Using efficient center-pivot 4D convolutions in a pyramidal architecture, the method gradually squeezes high-level semantic and low-level geometric cues of the hypercorrelation into precise segmentation masks in coarse-to-fine manner. The significant performance improvements on standard few-shot segmentation benchmarks of PASCAL-5i, COCO-20i, and FSS-1000 verify the efficacy of the proposed method.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Few-Shot Semantic Segmentation COCO-20i (1-shot) HSNet (ResNet-101) Mean IoU 41.2 # 51
FB-IoU 69.1 # 20
Few-Shot Semantic Segmentation COCO-20i (1-shot) HSNet (ResNet-50) Mean IoU 39.2 # 55
FB-IoU 68.2 # 23
Few-Shot Semantic Segmentation COCO-20i (5-shot) HSNet (ResNet-50) Mean IoU 46.9 # 47
FB-IoU 70.7 # 23
Few-Shot Semantic Segmentation COCO-20i (5-shot) HSNet (ResNet-101) Mean IoU 49.5 # 31
FB-IoU 72.4 # 15
Few-Shot Semantic Segmentation FSS-1000 (1-shot) HSNet (ResNet-50) Mean IoU 85.5 # 20
Few-Shot Semantic Segmentation FSS-1000 (1-shot) HSNet (ResNet-101) Mean IoU 86.5 # 16
Few-Shot Semantic Segmentation FSS-1000 (1-shot) HSNet (VGG-16) Mean IoU 82.3 # 22
Few-Shot Semantic Segmentation FSS-1000 (5-shot) HSNet (VGG-16) Mean IoU 85.8 # 19
Few-Shot Semantic Segmentation FSS-1000 (5-shot) HSNet (ResNet-101) Mean IoU 88.5 # 13
Few-Shot Semantic Segmentation FSS-1000 (5-shot) HSNet (ResNet-50) Mean IoU 87.8 # 17
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) HSNet (ResNet-50) Mean IoU 64.0 # 56
FB-IoU 76.7 # 32
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) HSNet (VGG-16) Mean IoU 59.7 # 76
FB-IoU 73.4 # 39
Few-Shot Semantic Segmentation PASCAL-5i (1-Shot) HSNet (ResNet-101) Mean IoU 66.2 # 34
FB-IoU 77.6 # 27
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) HSNet (ResNet-101) Mean IoU 70.4 # 31
FB-IoU 80.6 # 24
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) HSNet (ResNet-50) Mean IoU 69.5 # 39
FB-IoU 80.6 # 24
Few-Shot Semantic Segmentation PASCAL-5i (5-Shot) HSNet (VGG-16) Mean IoU 64.1 # 70
FB-IoU 76.6 # 37

Methods