SOTAVerified

Hypercorrelation Squeeze for Few-Shot Segmentation

2021-04-04Code Available1· sign in to hype

Juhong Min, Dahyun Kang, Minsu Cho

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

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.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
COCO-20i (1-shot)HSNet (ResNet-50)Mean IoU39.2Unverified
COCO-20i (1-shot)HSNet (ResNet-101)Mean IoU41.2Unverified
COCO-20i (5-shot)HSNet (ResNet-101)Mean IoU49.5Unverified
COCO-20i (5-shot)HSNet (ResNet-50)Mean IoU46.9Unverified
FSS-1000 (1-shot)HSNet (ResNet-50)Mean IoU85.5Unverified
FSS-1000 (1-shot)HSNet (ResNet-101)Mean IoU86.5Unverified
FSS-1000 (1-shot)HSNet (VGG-16)Mean IoU82.3Unverified
FSS-1000 (5-shot)HSNet (ResNet-50)Mean IoU87.8Unverified
FSS-1000 (5-shot)HSNet (ResNet-101)Mean IoU88.5Unverified
FSS-1000 (5-shot)HSNet (VGG-16)Mean IoU85.8Unverified
PASCAL-5i (1-Shot)HSNet (ResNet-101)Mean IoU66.2Unverified
PASCAL-5i (1-Shot)HSNet (ResNet-50)Mean IoU64Unverified
PASCAL-5i (1-Shot)HSNet (VGG-16)Mean IoU59.7Unverified
PASCAL-5i (5-Shot)HSNet (ResNet-101)Mean IoU70.4Unverified
PASCAL-5i (5-Shot)HSNet (ResNet-50)Mean IoU69.5Unverified
PASCAL-5i (5-Shot)HSNet (VGG-16)Mean IoU64.1Unverified

Reproductions