- Install required python packages from requirements.txt file.
- Place the input data images and json file as per samples in the repository.
- Path for Json file is defined in
config_2.py
file anddataset_500.json
contains relative paths for the input images. - Surrounding stream input size can be modified in
config_2.py
file through variablesimage_size_w_p
&image_size_h_p
, default values are 96,48 respectively. - Current data is splitted in 5 sets beforehand and Train:Validation:Test split is 2:1:2.
- Data pre-processing parallelization can be controlled through
max_workers
parameter insiamese_two_stream.py
, default = 10. suffix
andmodel_name
parameters fromsiamese_two_stream.py
can be modified to change the saved model names.
python siamese_two_stream.py train
python siamese_two_stream.py test models-car-96-96/
Test resuls on Vehicle Reid dataset
- Combined Results -
- Precision:94.56 | Recall:91.08 | F-score:92.62