HFA: Harmonic Feature Activation for few-shot semantic segmentation
Core Modules of HFA
Few-shot semantic segmentation aims at training a model which segments the novel class with few training data. We propose harmonic feature activation to effectively activate the target category in query image.
The Bilinear_Activation_slice module could be used for feature fusion.
The Semantic Diffusion module can be plugged before segmentation modules to refine the feature maps.
pytorch>=1.0
python>=3.6
numpy
pillow
opencv
PASCAL VOC dataset
MS COCO dataset
run main.ipynb
Please kindly refer to here
Part of our codes are based on the following repositories:
DEEPLAB-XCEPTION: https://github.com/jfzhang95/pytorch-deeplab-xception