This is a simplified implementation of the popular "Domain Separation Neural Network using Tensorflow. This work is just a humble attempt to simplify the official implementation see "https://github.com/tensorflow/models/tree/master/research/domain_adaptation", since there were several issues reported on the github and I faced them as well. After all it is our duty as engineers and future scientists to make people life easier and better, don't you think so? XD.
The main code parts including loss functions, encoders and the shared decoder were borrowed from the DSN official implementation. Big part of this implementation was inspired by the Domain Adaptation Neural Network official implementatio see "https://github.com/pumpikano".
Tested with TensorFlow=1.14.0 and Python 3.6.
The DSN.ipynb notebook implements the MNIST->MNISTM experiments mentioned in the paper. The domain classifier with reversal layer (DANN) is used here as a domain discrepancy loss since it was reported to be better than
Build MNIST-M dataset: MNIST-M dataset consists of MNIST digits blended with random color patches from the BSDS500 dataset. To generate a MNIST-M dataset, first download the BSDS500 dataset and run the create_mnistm.py script:
| Method | Target acc (paper) | Target acc (this repo w/ 30 epochs) |
|---|---|---|
| DSN | 0.832 | 0.821 |
It would be great to add other experiments to this repository. Feel free to make a PR if you decide to recreate other results from the papers or new experiments entirely.
In case you faced any issues or found any bugs feel free to contact me at "[email protected]"