thuml / MADA

Code release for "Multi-Adversarial Domain Adaptation" (AAAI 2018)

MADA

Code release for "Multi-Adversarial Domain Adaptation" (AAAI 2018)

Prerequisite

Protobuf Version 2.6.1

CUDA 7.5/8.0

Modification on Caffe

• Add "OuterProduct" layer to calculate weighted feature to input to each domain adversarial network;

Datasets

Office-31

The list files of Office-31 dataset are in office directory. You can download the original images here.

Training

The training command is as follows

./build/tools/caffe train -solver models/mada/solver.prototxt -weights ./models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel -gpu gpu_id

You need to download the reference_caffenet model here.

For different tasks, you need to set the "test_iter" parameter in the solver file as the size of the target dataset. For example, when "webcam" is used as target domain, you need to set the "test_iter" parameter as 795.

Contact

If you have any problem about our code, feel free to contact

• [e-mail]
• [e-mail]

or describe your problem in Issues.