Code release for "Multi-Adversarial Domain Adaptation" (AAAI 2018)
Protobuf Version 2.6.1
CUDA 7.5/8.0
- Add "OuterProduct" layer to calculate weighted feature to input to each domain adversarial network;
The list files of Office-31 dataset are in office directory. You can download the original images here.
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.
If you have any problem about our code, feel free to contact
or describe your problem in Issues.