This is the source code of MSMNN. It is built on top of the SCAN (https://github.com/kuanghuei/SCAN) in PyTorch. In our experiments, we use two NVIDIA Tesla K80 GPUs for the parallel training.
- python == 2.7
- PyTorch 0.3.0
- numpy >= 1.12.1
Download the dataset files. We use the visual local features created by SCAN, downloaded here, and salience features can be extracted by R^3 net (https://github.com/zijundeng/R3Net). After extracting the salience features, we need to connect local features and salience features by the order of images.
python train.py
python evalution1.py