In the image, there is a large, beautifully decorated poster hanging on a wall. The poster appears to be advertising a product or service, as it features a vibrant and eye-catching design, including a striking color palette and bold text. Below the poster, there is another document displayed on the wall, which appears to be a detailed explanation or testimonial about the product or service being advertised on the poster. The document is likely intended to provide more information or highlight the benefits of the product to potential customers, as it features a clear and concise layout, with text organized into concise paragraphs and bullet points. Overall, the image presents a well-designed and informative advertising display, aimed at capturing the attention of passersby and promoting the product or service being showcased.
Text transcribed from the image:
Continual Compatible Representation for Re-indexing Free Lifelong Person Re-ident
Zhenyu Cui¹, Jiahuan Zhou1, Xun Wang2, Manyu Zhu², Yuxin Peng1*
1Wangxuan Institute of Computer Technology, Peking University 2ByteDance Inc
ReID): Aims to learn from
a person across different
orical images in the gallery
or testing (Re-indexing).
Data
Privacy
ReID w/o Re-indexing
New Model
QQ
ID:3
ID:5
Old Gallery Features
Query Features
when raw images in the
acy concerns, resulting in
and the gallery features
on L-ReID performance.
100
Continual
Compatible
Representation
New Model
ID:3 ID:5
Old Gallery Features
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.
➤ 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.
(a) Training at s Stage
Old data
New data
Old Model
New Mode
(b) Transfer at s Stage
F
ID:1 ID:2
Updated Features
Query Features
Continual Compatible Transfer (CCT)
Balanced Compatible Distillation (BCD)
Multiply
20
Forward Mapping Module (FMM))
Softmax
Sum
Balanced Anti-forgetting Distillation (BAD)
MLP
H
FC
Knowledge Capturing Module (KCM)
Multiply
00-1-1-0000-1
ID:
New Gallery Feature Set
ID:
ID:
Old Gallery
Feature Set
Transferred
Feature Set
Experiments:
The experiment o
superiority of our pr
Mark
Task
Method
mAE
Joint Train
68.1
SPD [28]
35.0
LWF [11]
56.
CRL [40]
58.
L-ReID
AKA [18]
58
MEGE [20] 39
RFL-ReID
PatchKD [27] 68
Ours 69
LwF* [11] 39
AKA* [18] 3-
CVS* [30] 3
PatchKD* [27] 6
Ours
Our C2R achieve
Conclusion
We propose
tackling the d
• Two balance
forgetting of
•
H
(c) Testing at s Stage
ID: N
ID: N+1 ID: M
ID: 1
•
Extensive e
Extracted
Features:
ID: 182
Query
demonstrate
m on L-ReID task, called
tification (RFL-ReID).
ntation (C2R) scheme to
d gallery in L-ReID.
T) and two balanced
eve L-ReID without re-
Pipeline:
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.
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 data.
More
Information
Go!
Related
Works
Go!
回回
[Lab
[1] Zhenyu
Rectificatio
[2] Zhenyu
Re-Identific