SOTAVerified

HM: Hybrid Masking for Few-Shot Segmentation

2022-03-24Code Available0· sign in to hype

Seonghyeon Moon, Samuel S. Sohn, Honglu Zhou, Sejong Yoon, Vladimir Pavlovic, Muhammad Haris Khan, Mubbasir Kapadia

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

We study few-shot semantic segmentation that aims to segment a target object from a query image when provided with a few annotated support images of the target class. Several recent methods resort to a feature masking (FM) technique to discard irrelevant feature activations which eventually facilitates the reliable prediction of segmentation mask. A fundamental limitation of FM is the inability to preserve the fine-grained spatial details that affect the accuracy of segmentation mask, especially for small target objects. In this paper, we develop a simple, effective, and efficient approach to enhance feature masking (FM). We dub the enhanced FM as hybrid masking (HM). Specifically, we compensate for the loss of fine-grained spatial details in FM technique by investigating and leveraging a complementary basic input masking method. Experiments have been conducted on three publicly available benchmarks with strong few-shot segmentation (FSS) baselines. We empirically show improved performance against the current state-of-the-art methods by visible margins across different benchmarks. Our code and trained models are available at: https://github.com/moonsh/HM-Hybrid-Masking

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
COCO-20i (1-shot)HSNet (HM, ResNet-50)Mean IoU44.3Unverified
COCO-20i (1-shot)ASNet (HM, ResNet-50)Mean IoU44.7Unverified
COCO-20i (1-shot)ASNet (HM, ResNet-101)Mean IoU45.9Unverified
COCO-20i (1-shot)HSNet (HM, ResNet-101)Mean IoU46.5Unverified
COCO-20i (1-shot)VAT (HM, ResNet-50)Mean IoU43.2Unverified
COCO-20i (5-shot)ASNet (HM, ResNet-101)Mean IoU50.6Unverified
COCO-20i (5-shot)HSNet (HM, ResNet-101)Mean IoU50.6Unverified
COCO-20i (5-shot)HSNet (HM, ResNet-50)Mean IoU49.4Unverified
COCO-20i (5-shot)ASNet (HM, ResNet-50)Mean IoU48.4Unverified
COCO-20i (5-shot)VAT (HM, ResNet-50)Mean IoU48.3Unverified
COCO-20i -> Pascal VOC (1-shot)HSNet (HM, ResNet-101)Mean IoU66.5Unverified
COCO-20i -> Pascal VOC (1-shot)HSNet (HM, ResNet-50)Mean IoU65.2Unverified
COCO-20i -> Pascal VOC (1-shot)VAT (HM, ResNet-50)Mean IoU65.1Unverified
COCO-20i -> Pascal VOC (5-shot)VAT (HM, ResNet-50)Mean IoU69.7Unverified
COCO-20i -> Pascal VOC (5-shot)HSNet (HM, ResNet-50)Mean IoU69.7Unverified
COCO-20i -> Pascal VOC (5-shot)HSNet (HM, ResNet-101)Mean IoU70.9Unverified
FSS-1000 (1-shot)HSNet (HM, ResNet-50)Mean IoU87.1Unverified
FSS-1000 (1-shot)VAT (HM, ResNet-101)Mean IoU90.2Unverified
FSS-1000 (1-shot)VAT (HM, ResNet-50)Mean IoU89.4Unverified
FSS-1000 (1-shot)HSNet (HM, ResNet-101)Mean IoU87.8Unverified
FSS-1000 (5-shot)HSNet (HM, ResNet-50)Mean IoU88Unverified
FSS-1000 (5-shot)VAT (HM, ResNet-101)Mean IoU90.5Unverified
FSS-1000 (5-shot)VAT (HM, ResNet-50)Mean IoU89.9Unverified
FSS-1000 (5-shot)HSNet (HM, ResNet-101)Mean IoU88.5Unverified
PASCAL-5i (1-Shot)HSNet (HM, ResNet-50)Mean IoU65Unverified
PASCAL-5i (1-Shot)VAT (HM, ResNet-50)Mean IoU65.8Unverified
PASCAL-5i (1-Shot)HSNet (HM, ResNet-101)Mean IoU66.7Unverified
PASCAL-5i (1-Shot)VAT (HM, ResNet-101)Mean IoU67.8Unverified
PASCAL-5i (5-Shot)HSNet (HM, ResNet-50)Mean IoU67.1Unverified
PASCAL-5i (5-Shot)VAT (HM, ResNet-50)Mean IoU68.2Unverified
PASCAL-5i (5-Shot)HSNet (HM, ResNet-101)Mean IoU69.3Unverified
PASCAL-5i (5-Shot)VAT (HM, ResNet-101)Mean IoU70.9Unverified

Reproductions