This is the implementation of the good work PFENet: Prior Guided Feature Enrichment Network for Few-shot Segmentation. Based on this paper, we rewrite the code for higher training efficiency and better code reuse. Thanks to the official implementation.
In this paper, PASCAL-5i and COCO 2014 are used for evaluation.
We have preprocessed PASCAL-5i, and you can download (pass code: b74i) it directly.
All datasets are stored in DocSet (*.ds) files, since this file format is portable and efficient. To support ds files in your system, install the "dcoset" package for your python environment.
pip3 install docsetpython3 train.py --data-path /path/of/pascal5i.ds --gpu 0 --batch-size 8 --num-epochs 30