Occluded Person Re-identification via Saliency-Guided Patch Transfer

Authors: Lei Tan, Jiaer Xia, Wenfeng Liu, Pingyang Dai, Yongjian Wu, Liujuan Cao

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental evaluations conducted on occluded and holistic person re-identification benchmarks demonstrate that SPT provides a significant performance gain among different Vi T-based Re ID algorithms on occluded Re ID.
Researcher Affiliation Collaboration 1Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China. 2Tencent Youtu Lab, China.
Pseudocode No The paper does not contain a clearly labeled pseudocode or algorithm block. Figure 2 provides a framework overview, but not algorithmic steps.
Open Source Code No The paper does not provide any statement about making its source code publicly available or a link to a code repository.
Open Datasets Yes Occluded-Duke (Miao et al. 2019) is a large-scale dataset for occluded person re-identification. The training set consists of 15,618 images of 702 persons... Market-1501 (Zheng et al. 2015) is a holistic Re ID dataset... Duke MTMC-re ID (Zheng, Zheng, and Yang 2017) contains 36,441 images...
Dataset Splits Yes Occluded-Duke (Miao et al. 2019) is a large-scale dataset for occluded person re-identification. The training set consists of 15,618 images of 702 persons. The query set consists of 2,210 images of 519 persons and the gallery set consists of 17,661 images of 1,110 persons.
Hardware Specification Yes All experiments are implemented using Py Torch on a single Nvidia 3090 Ti.
Software Dependencies No The paper mentions "Py Torch" as the implementation framework but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes We resize all input images to 256 128 and employ commonly used data augmentation techniques, including horizontal flipping, padding, and random cropping. During the training phase, we train the SPS in the first 50 epochs and then freeze the SPS to train the network for 120 epochs. Each mini-batch with 64 images, including 16 identities and 4 images per identity. We employ the SGD optimizer and initialize the learning rate as 0.008 with cosine learning rate decay. We set β as 0.3 and 0.5 when training for occluded-Duke and occluded-Re ID, respectively. For controlling the minimized OIo U, we set α1 as 0.5. Regarding α2, we set it as 0.1 and apply a ranking strategy during each batch.