Two individuals are deeply engaged in reading and discussing a research poster at an academic conference or symposium. The poster is titled "Learning Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification" and appears to present a study conducted by researchers from Peking University and ByteDance Inc. Detailed diagrams, charts, and textual explanations are visible, indicating the complexity and depth of the research. The environment suggests a professional setting with modern, industrial-style ceiling lights and additional posters or displays in the background. The individuals, appearing to be attendees or fellow researchers, are giving focused attention to the material, possibly contemplating the implications and methodologies discussed.
Text transcribed from the image:
Featur
Learning Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification
Zhenyu Cui¹, Jiahuan Zhou, Xun Wang2, Manyu Zhu, Yuxin Peng1*
1Wangxuan Institute of Computer Technology, Peking University 2Byte Dance Inc
(L-ReID): Aims to learn from
a person across different
orical images in the gallery
testing (Re-indexing).
xing
Method:
Overview: A Continual Compatible Representation (C2R) scheme is
proposed for RFL-ReID to continuously update old features in the
gallery to make it compatible with new query features.
space!
Continual Compatible Transfer (CCT): Update old gallery
features continuously and transfer old feature to the new feature
space, adaptively capturing the knowledge from different domains.
Balanced Compatible Distillation (BCD): Achieve the
compatibility between the transferred features and the new ones,
preserving the relationship between the old and the transferred
features in a unified feature space.
Balanced Anti-forgetting Distillation (BAD): Eliminate the
accumulated forgetting of old knowledge during the continuous
transfer, balancing the old and the new discriminative information.
Canal Compatible Transfer CCT
Experiments:
The experiment on the benchmark L-R
superiority of our proposed C2R.
Method
Market-1501 CUHK-SYSU DeMTMC-ReD
RUI
MAP
Joint Train
SPD 125)
68.1 852 814
356 612 61.7
83.3 604
LwF[i]
563 77.1 729
296
CRL 140
580
782
72.5
75.1
28.3
L-RelD
AKA (1) 58.1
74.
28.7
390 616
733
76.6 165
RFL-RelD
MEGE [201
PatchKD 1271 685 85.7 75.6 28
69.0 36.X
76,7 "
LF)
580 400
39.1
522
36.1
AKA [18]
366
38% 55.6 49.0
CVS [301
784 37
PatchKD (271) 614
Ours 627 79.7
Our C2R achieves stable
anced
out re-
Hin
Conclusion
We propose C2R for the pra
tackling the data privacy issu
Two balanced distillation
forgetting of old knowledge
Extensive experiments
demonstrate the effectiven
L
Pipeline:
More
Training Stage: CCT, BCD, and BAD are jointly optimized.
Transferring Stage: CCT network is employed to update the old
feature set after each training stage in L-ReID.
Information
Go!
Testing Stage: Query features calculated by the new model is
used to directly match the updated gallery features to achieve re
indexing free lifelong person re-identification.
Three stages are executed sequentially when facing new dat
Related