Skip to content

Latest commit

 

History

History
27 lines (27 loc) · 819 Bytes

README.md

File metadata and controls

27 lines (27 loc) · 819 Bytes

PIPA baseline

This is an simplified implmentation for the paper Learning Deep Features via Congenerous Cosine Loss for Person Reconition
It has only implemented softmax loss.
The coco loss version may be released some day.

Requirement

  • Python 2.7
  • MXNET 1.3
  • numpy
  • matplotlib (not necessary unless the need for the result figure)

Network

The backbone of the network is Inception pretrained in ImageNet.
You can specify the network by the param --network.

Train & Test

Train on head

sh run_head.sh

Train on face

sh run_face.sh

Train on the whole body

sh run_person.sh

References

Y. Liu, H. Li, and X. Wang. Learning deep features via congenerous cosine loss for person recognition. CoRR,abs/1702.06890, 2017.