This is the code of our ICME 2018 paper, "RAM: A Region-Aware Deep Model for Vehicle Re-Identification".
If you find this help, please kindly cite our paper:

@inproceedings{cime-ram-liu,
  Author = {Liu, Xiaobin and Zhang, Shiliang and Huang, Qingming and Gao, Wen},<br>
  Booktitle = {ICME},
  Title = {RAM: A Region-Aware Deep Model for Vehicle Re-Identification},
  Year = {2018}
}

You can simply train a RAM on VeRi by running:

sh train_veri.sh

The final model is saved as "snapshot/veri-RAM-finetune_iter_60000.caffemodel". Our models and extracted features on VeRi can be downloaded from here.

We provide a new caffe layer to sample mini-batch. Please refer to "prototxt/train_RAM.prototxt" for an example of usage.

We provide an evaluate script, "evaluate.m", on VeRi following https://github.com/VehicleReId/VeRidataset

We provide a tools in caffe to extract features and write features to binary files. We also provide a tools to read features from binary file, "read_code.m", and a tools to normalize features, "norm_code.m". An example of the usage of these tools can be found in "evaluate.m".

Email: [email protected]
Homepage: https://liu-xb.github.io
Please feel free to contact me if you have any question.