LIRENDA621/MMSports2023_-Player_Reidentification_Challenge

[Paper][Challenge]

Please note that the MMSports 2022 championship scheme is our baseline and our code is built based on this [repository].

Code for the paper: Exploring Loss Function and Rank Fusion for Enhanced Person Re-identification

Setup environment

Clone the repo:

git clone https://github.com/LIRENDA621/MMSports2023_-Player_Reidentification_Challenge.git
cd MMSports2023_-Player_Reidentificatio_Challenge

Create conda environment:

pip install -r requirements.txt

Approach

Usage

Steps for Training and Evaluation:

1. get data: download_data.py

2. create DataFrames: preprocess_data.py

3. training:

   train_our_post.py

   The main training script will output two post-processing results during each verification, one is last year's championship solution, and the other is our implementation. The differences can be found in the paper.

   train_our_post_resnet.py

   This training script facilitates the use of the ResNet model to conduct relevant experiments in the paper.

4. evaluation: evaluate.py

5. final predictions:

   predict.py

   predict_fortestset.py

   The script is convenient for performing ablation experiments on the test set and does not need to be used.

Model Fuse

1. similarity fuse: my_py/sim_fuse.py

2. dissimilarity fuse: my_py/dissim_fuse.py

In our solution, we first perform similarity fusion and dissimilarity fusion on multiple models respectively, and then perform similarity fusion on the two results.

All settings are done by the configuration dataclass at the beginning of the scripts.

download_data.py downloads and unzips the challenge data from the provided challenge toolkit.