search

QizaoWang / FIRe-CCReID

Exploring Fine-Grained Representation and Recomposition for Cloth-Changing Person Re-Identification [TIFS 2024]
https://arxiv.org/abs/2308.10692
11 stars 1 forks source link

The results after training differ from those reported in the paper #2

Open Crush52-sys opened 1 week ago

Crush52-sys commented 1 week ago

Hello, after following the training commands in the readme, I obtained the following results: For the LTCC dataset: Best mAP is 17.6509%, and Best Rank-1 is 38.7755%. For the PRCC dataset: Best mAP is 58.4294%, and Best Rank-1 is 58.9896%.

Could you kindly advise if there are any parameters that need adjustment?

QizaoWang commented 1 week ago

Thank you for your interest in our work! The results reported in our paper were achieved without performing grid tuning on the hyperparameters. Adjusting hyperparameters, such as --eps and --FAR_weight, for different datasets may lead to improved performance, while you can achieve similar results with the released code. Variations in training results could also be influenced by factors like randomness and differences in hardware. For reference, the experiments in our paper were conducted on NVIDIA GeForce GTX 1080 Ti. Additionally, we recommend trying larger datasets, such as DeepChange and LaST, which tend to produce more stable results.