Multiple Granularity Network

Reproduction of paper: Learning Discriminative Features with Multiple Granularities for Person Re-Identification

About

This is a non-official pytorch re-production of paper: Learning Discriminative Features with Multiple Granularities for Person Re-Identification. Still Work In Progress.

Please cite and refer to:

@ARTICLE{2018arXiv180401438W,
    author = {{Wang}, G. and {Yuan}, Y. and {Chen}, X. and {Li}, J. and {Zhou}, X.},
    title = "{Learning Discriminative Features with Multiple Granularities for Person Re-Identification}",
    journal = {ArXiv e-prints},
    archivePrefix = "arXiv",
    eprint = {1804.01438},
    primaryClass = "cs.CV",
    keywords = {Computer Science - Computer Vision and Pattern Recognition},
    year = 2018,
    month = apr,
    adsurl = {http://adsabs.harvard.edu/abs/2018arXiv180401438W},
    adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

Code is only tested against python 2.7 and pytorch 0.4.

Implementation

• mgn/mgn.py: re-production of Multiple Granularity Network.

• mgn/ide.py: baseline ResNet-50 based model, which is a rewritten from Person reID baseline pytorch.

• mgn/triplet.py: triplet semi-hard sample mining loss.

• mgn/market1501.py: Market-1501 dataset.

• Market-1501-v15.09.15/: Market-1501 dataset root directory.

Current Progress

• 2018-04-28: mAP=0.579464, r@1=0.798694, r@5=0.909739, r@10=0.938539