- Python == 3.6 (conda)
- NVIDIA driver version: 460.x
- NVIDIA CUDA version: 11.2
- NVIDIA cuDNN version: 8.1.0
$ apt-get install ffmpeg
- RAM size: >= 12 GB;
- Hard disk free space: >= 50 GB;
.
├── data # Directory with the dataset
| └── ground_truth # Wisenet Ground Truth Files
| └── output # Report tracking of person re-identification
| └── wisenet_dataset # Dataset
├── libs # External libs
| └── facenet # https://github.com/davidsandberg/facenet
├── notebooks # Notebooks codes
├── output # Results of metrics evaluation
├── scripts # Shell scripts for run project
├── src # Source code
The person re-identification project is composed of two face recognition neural networks, Facenet and Mobilnet. To run the project, you will need two different environments,
$ apt-get install ffmpeg
$ conda create -n tcc-25 python=3.6 anaconda
$ conda activate tcc-25
$ conda install -c anaconda cudatoolkit==11.2 -y
$ conda install -c anaconda cudnn -y
$ pip install -r requirements_mobilenet.txt
$ conda create -n tcc-17 python=3.6 anaconda
$ conda activate tcc-17
$ conda install -c anaconda cudatoolkit==9.0 -y
$ conda install -c anaconda cudnn -y
$ pip install -r requirements_facenet.txt
- Download Wisenet dataset, in official repository
.
├── data # Directory with the dataset
| └── wisenet_dataset # Dataset
| | └── video # Directory with sets
| | | └── set_1 # Set 1
| | | └── set_2 # Set 2
| | | └── set_3 # Set 3
| | | └── set_4 # Set 4
-
Run the script that will convert all videos into frames and save only those that have a face detected by MTCNN. Or make download of images used in paper.
$ cd scripts
$ get_faces.sh
.
├── data # Directory with the dataset
| └── wisenet_dataset # Dataset
| | └── videos_frames
| | | └── videio1_1 # Set 1 with camera 1
| | | └── videio1_2 # Set 1 with camera 2
| | | └── videio1_3 # Set 1 with camera 3
| | | └── videio1_4 # Set 1 with camera 4
| | | └── videio1_5 # Set 1 with camera 5
Noted: Use the FaceNet environment
- In directory
data/wisenet_dataset/videos_frames, the human faces resulting from theget_faces.shscript will be saved, make the separation of the faces by user, being that unknown user will be calledUNK. Or make download of images used in paper
.
├── data # Directory with the dataset
| └── wisenet_dataset # Dataset
| | └── database_frames # Directory faces for trainig model
| | | └── ID1 # Faces of person 1
| | | └── ID2 # Faces of person 2
| | | └── ID3 # Faces of person 3
| | | └── UNK # Faces of person unknown
Noted: Select only images from videos that will not be used in the experiment.
- Run notebook
notebooks/train/train_mobilenet.ipynbfor training mobileNet model, or download the models pre-trained e labels
.
├── models # Directory with the models
| └── mobilenet
| | └── assets
| | └── variables
| | └── keras_metada.pb
| | └── saved_model.pb
| └── labels.txt
Noted: Use the MobileNet environment
- Download VGGFace2 pre-trained model trained by David Sandberg.
.
├── models # Directory with the models
| └── facenet
| | └── 20180402-114759.pb
| | └── model-20180402-114759.ckpt-275.data-00000-of-00001
| | └── model-20180402-114759.ckpt-275.index
| | └── model-20180402-114759.meta
- Run notebook
notebooks/train/train_facenet.ipynbfor training FaceNet model, or download the models pre-trained.
.
├── models # Directory with the models
| └── facenet
| | └── 20180402-114759.pb
| | └── model-20180402-114759.ckpt-275.data-00000-of-00001
| | └── model-20180402-114759.ckpt-275.index
| | └── model-20180402-114759.meta
| | └── one_shot_classifier.pkl
Noted: Use the FaceNet environment
-
Run the script that will do the person re-identification used MobileNet.
$ cd scripts
$ multi_mobilenet.sh
number_of_set_videosExample: $ multi_mobilenet.sh 1
.
├── data
| └── output
| | └── set_1
| | | └── tracking_predict_db_mobilenet.json
Noted: Use the MobileNet environment
-
Run the script that will do the person re-identification used FaceNet.
$ cd scripts
$ multi_facenet.sh
number_of_set_videosExample: $ multi_facenet.sh 1
.
├── data
| └── output
| | └── set_1
| | | └── tracking_predict_db_facenet.json
Noted: Use the FaceNet environment
-
Execute metrics evaluate of MobileNet report tracking.
$ cd scripts
$ evaluate_mobilenet.sh
number_of_set_videosExample: $ evaluate_mobilenet.sh 1
.
├── output
| └── set_1
| | └── mobilenet
| | | └── metrics.csv
> Noted: Use the MobileNet environment
- Execute metrics evaluate of FaceNet report tracking.
$ cd scripts
$ evaluate_facenet.sh
number_of_set_videos
Example: $ evaluate_facenet.sh 1
.
├── output
| └── set_1
| | └── facenet
| | | └── metrics.csv
Noted: Use the MobileNet environment
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent