abrams90/vehicle-retrieval-kcnns

vehicle image retrieval using k CNNs ensemble method

Vehicle Retrieval

Multi-CNN feature ensemble method.

1. We have trained 4 GoogLeNet with different task, e.g, triplet loss (full image), vehicle_id softmax loss (full image), vehicle_id softmax loss (upper half image) and model_id softmax loss (lower half image).
2. In inference stage, we concat the pool5/7x7_s1 layer feature from the 4 GoogLeNet together. Finally, return the k nearest vehicles.

Competition & Dataset

We used the method in Vehicle Retrieval task of The 3rd National Gradute Contest on Smart-CIty Technology and Creative Design, China. We ranked 1st and won the special prize in the final!

The Dataset used in Vehicle Retrieval task: PKU VehicleID. Note: if you want to use the dataset, go to the website and ask for the download link.

Platform

• CPU or GPU: CPU only
• OS: Windows x64
• DL tool: Caffe
• Compiler: VS2013

Usage

1. Windows Caffe (use this version) setup.
2. Download or git clone the current project.
3. Copy or move vs_vehicle_retrieval_kCNNs folder into caffe/windows and add vehicle_retrieval_kCNNs.vcxproj project into Caffe solution in VS2013, compile it with Release mode.
4. Modify run.bat, mainly set the path. Finally, run run.bat in cmd, you'll get a xml result file.

---

车辆精确检索

第三届全国研究生智慧城市技术与创意设计大赛车辆精确检索任务第一名，总决赛特等奖。

数据集：PKU VehicleID

方法

基于深度学习的多模型集成方法。

平台

CPU，Windows系统，Caffe，VS2013

使用

1. 下载、配置、编译Caffe官方windows版(https://github.com/BVLC/caffe/tree/714d0acad8c66d64ddf7b83b9a239f7efc017894)
2. 下载本工程
3. 将文件夹vs_vehicle_retrieval_kCNNs复制到caffe/windows目录下，并在vs中把vehicle_retrieval_kCNNs.vcxproj项目添加到Caffe解决方案下，使用Release模式编译生成可执行文件。
4. 修改run.bat中的路径，运行它，即可得到实验结果。