Statistically-motivated Second-order Pooling (ECCV2018)

Code for our work on ECCV2018. Paper URL: https://arxiv.org/abs/1801.07492

It is implemented in Keras with Tensorflow as backend.

Requirements:

• Tensorflow: 1.4.0
• Keras 2.1.2

Notes

The first version of the code is put in snapshot folder, which containing only the implementation of SMSOP structure. You can obtain these by calling get_cov_block(option) function in the main.py.

def get_cov_block(cov_branch):
    if cov_branch == 'smsop':
        covariance_block = covariance_block_newn_wv
    elif cov_branch == "smsop-equ":
        covariance_block = covariance_block_pv_equivelent
    else:
        raise ValueError('covariance cov_mode not supported')

    return covariance_block