PMGN: Prototype-Matching Graph Network for Heterogeneous Domain Adaptation

This is the official implementation of MM'20 paper 'Prototype-Matching Graph Network for Heterogeneous Domain Adaptation'

Please install the required libaries first, and then run code

python train.py

Source domain	Target domain	Extra Parameter Setting	Result (Pytorch 1.6.0)
Amazon_Surf	Webcam_Decaf	--dis_loss=0.15	98.62
rand_GR	rand_SP_10		99.60
NUSTAG	IMGNET		86.29

Full dataset can be download at:

• Office-Caltech: https://www.dropbox.com/s/i87qac7kzkwy8d9/datasets.zip

• NUS-IMG: https://www.dropbox.com/s/i87qac7kzkwy8d9/datasets.zip

  • (Credits to the Transfer Nerual Trees Repo https://github.com/wyharveychen/TransferNeuralTrees)

• Multilingual Reuters Collection: https://drive.google.com/drive/folders/1OeK9wJRjIvxt-tRNZw9cNXAuiytTncu6?usp=sharing

Working Directory

PMGN
|-data
|- mrc
|- Office-Caltech
| - ...
|- utils
|- ...
|- train.py