So far only evaluation code is supported. Run eval.py script to evaluate a model on a particular subset.
You will need to use lua to convert dataset to the compatible format. eval.py has functional definitions
for all models, so that you can easily grab the parameters and model function and reuse it in your own code.
The numbers on Brown dataset are slightly different due to different evaluation code:
| sets | siam2stream_l2 | siam2stream | 2ch | siam | siam_l2 | 2ch2stream |
|---|---|---|---|---|---|---|
| yosemite, notredame | 5.63 | 5.37 | 3.05 | 5.76 | 8.40 | 2.22 |
| yosemite, liberty | 12.02 | 11.03 | 9.02 | 13.58 | 18.90 | 7.48 |
| notredame, yosemite | 12.85 | 10.29 | 5.73 | 12.63 | 15.18 | 3.99 |
| notredame, liberty | 7.93 | 6.19 | 5.86 | 8.62 | 12.55 | 5.46 |
| liberty, yosemite | 13.32 | 9.42 | 7.65 | 15.29 | 20.11 | 5.27 |
| liberty, notredame | 5.19 | 3.16 | 3.02 | 4.43 | 6.05 | 1.88 |
Install PyTorch following instructions from https://pytorch.org, then run:
pip install -r requirements.txtAnd install torchnet:
pip install git+https://github.com/pytorch/tnt.git@master