In this project, we used deep learning method for attributes classification and re-identification tasks. For each task, we tried to combine several methods to improve performance and also experiment un different network architectures.
code:
- re_id.ipynb : using to train network with triplet loss
- test_re_id_final.ipynb : using to generate final result with reranking task
- test_re_id_map.ipynb: using map metric to evaluate different methods
- training.ipynb: training file
- prepare_data.ipynb:
- Data augmentation
- split training and validation set
- classification_evaluate.ipynb: evaluate classification task
- evaluate different models on validation set
- generate classification result
- identification_evaluate.ipynb: build embedding vector and evaluate re-identification task
- model.py: models for classification task
result:
- reid test.txt : result for re-identification task
- classification_test.csv : result for classification task