Person_reID_baseline_matconvnet

Matconvnet implement of Person re-identification baseline. We arrived Rank@1=87.74% mAP=69.46% only with softmax loss.

Installation

1. Clone this repo

   git clone https://github.com/layumi/Person_reID_baseline_matconvnet.git
   cd Person_reID_baseline_matconvnet
   mkdir data

2. Compile matconvnet

   You just need to uncomment and modify some lines in gpu_compile.m and run it in Matlab. Try it~

   If you fail in compilation, you may refer to http://www.vlfeat.org/matconvnet/install/

Train

1. Add your dataset path into prepare_data.m and run it. Make sure the code outputs the right image path.

2. Download the model pre-trained on ImageNet. And put the model into './data'.

(bash) wget http://www.vlfeat.org/matconvnet/models/imagenet-resnet-50-dag.mat

3. Run train_id_net_res_market_baseline.m.

Test

1. After training, you can run test/test_gallery_res_crazy_pool5.m to extract the features of images in the gallery and query set. They will store in a .mat file. Then you can use it to do evaluation.

2. run baseline/zzd_evaluation_res_faster.m to get the final result.