search

human-analysis / intrinsic-dimensionality

PyTorch Implementation of CVPR'19 - On the Intrinsic Dimensionality of Image Representation
http://hal.cse.msu.edu
23 stars 5 forks source link

Table 2 dose not meet my reproduce. #4

Closed TJJTJJTJJ closed 5 years ago

TJJTJJTJJ commented 5 years ago

How do you get the feature of LFW? In your table 2, 512-dim only achieve 96.74\%, but when I look at the implement of sphereface, it could achieve 99.30\%. And when I test the LFW feature the sphereface released, it could achieve 98.02\%, and, when I use PCA to reduce the dimension to 16, it still achieve 0.8481, which is higher than your reported 32.67\%.

So I have a question: where does your feature come from? Could you give me your lfw feature? I want to reproduce the table 2.

vboddeti commented 5 years ago

The result of 99.30% on LFW with Sphereface is probably from the standard LFW evaluation protocol. In our paper we instead follow the BLUFR evaluation protocol on LFW. On the BLUFR protocol Sphereface achieves 96.74% only.