Camera-Conditioned Stable Feature Generation for Isolated Camera Supervised Person Re-IDentification

Wu, Chao; Ge, Wenhang; Wu, Ancong; Chang, Xiaobin

doi:10.1109/cvpr52688.2022.01960

Cited by 15 publications

(13 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…al [43] design a multi-camera negative loss that ensures the most similar negative image is found in other cameras. CCFP [9] and CCSFG [38] try to generate synthetic features to complement the lack of cross-camera positive images. In specific, CCFP [9] utilizes camera-specific batch normalization, while CCSFG [38] adopts camera-conditioned variational encoder to synthesize crosscamera positive features.…”

Section: Isolated-camera Supervised Re-idmentioning

confidence: 99%

“…CCFP [9] and CCSFG [38] try to generate synthetic features to complement the lack of cross-camera positive images. In specific, CCFP [9] utilizes camera-specific batch normalization, while CCSFG [38] adopts camera-conditioned variational encoder to synthesize crosscamera positive features. After obtaining the synthetic features, they compute a consistency loss to minimize the distance of those cross-camera positive pairs.…”

Section: Isolated-camera Supervised Re-idmentioning

confidence: 99%

“…Therefore, it would be desirable if a re-ID model can learn from limited training data given only single-camera images of each person. This specific task is referred to as "Isolated Camera Supervised (ISCS) re-ID", and has been studied in a few works [9,38,43] previously.…”

Section: Introductionmentioning

confidence: 99%

“…The resulting model learns to discriminate cross-camera images with less camera bias involved. Later works CCFP [9] and CCSFG [38] both resort to cross-camera feature generation to make up for the absence of cross-camera positive images. CCFP adopts cameraspecific BNs [18], while CCSFG [38] proposes camera-conditioned variational auto-encoder to generate cross-camera fake positive features.…”

Section: Introductionmentioning

confidence: 99%

“…Later works CCFP [9] and CCSFG [38] both resort to cross-camera feature generation to make up for the absence of cross-camera positive images. CCFP adopts cameraspecific BNs [18], while CCSFG [38] proposes camera-conditioned variational auto-encoder to generate cross-camera fake positive features. Nevertheless, the generated feature may be semanticallynoisy or lack diversity in pedestrian view or motion, hindering model performance from being further improved.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Learning Intra and Inter-Camera Invariance for Isolated Camera Supervised Person Re-identification

Wang,

Gong

2023

Proceedings of the 31st ACM International Conference on Multimedia

View full text Add to dashboard Cite

Supervised person re-identification assumes that a person has images captured under multiple cameras. However when cameras are placed in distance, a person rarely appears in more than one camera. This paper thus studies person re-ID under such isolated camera supervised (ISCS) setting. Instead of trying to generate fake cross-camera features like previous methods, we explore a novel perspective by making efficient use of the variation in training data. Under ISCS setting, a person only has limited images from a single camera, so the camera bias becomes a critical issue confounding ID discrimination. Cross-camera images are prone to being recognized as different IDs simply by camera style. To eliminate the confounding effect of camera bias, we propose to learn both intraand inter-camera invariance under a unified framework. First, we construct style-consistent environments via clustering, and perform prototypical contrastive learning within each environment. Meanwhile, strongly augmented images are contrasted with original prototypes to enforce intra-camera augmentation invariance. For inter-camera invariance, we further design a much improved variant of multi-camera negative loss that optimizes the distance of multi-level negatives. The resulting model learns to be invariant to both subtle and severe style variation within and cross-camera. On multiple benchmarks, we conduct extensive experiments and validate the effectiveness and superiority of the proposed method. Code will be available at https://github.com/Terminator8758/IICI. CCS CONCEPTS• Information systems → Top-k retrieval in databases.

show abstract