Visible-infrared person re-identification (VI-ReID) is a cross-modality retrieval problem, which aims at matching the same pedestrian between visible and infrared cameras. Due to existence of pose variation, occlusion, huge visual differences two modalities, previous studies mainly focus on learning image-level shared features. Since they usually learn global representation or extract uniformly divided part features, these methods are sensitive misalignments. In this paper, we propose...

Matching the daytime visible and nighttime infrared person images, namely re-identification (VI-ReID), is a challenging cross-modality retrieval problem. Due to difficulty of data collection annotation in surveillance, VI-ReID usually suffers from noise problems, making it directly learn part discriminative features. In order improve discriminability enhance robustness against noisy this paper proposes novel dynamic tri-level relation mining (DTRM) framework by simultaneously exploring...

This paper introduces a simple yet powerful channel augmentation for visible-infrared re-identification. Most existing operations designed single-modality visible images do not fully consider the imagery properties in to infrared matching. Our basic idea is homogeneously generate color-irrelevant by randomly exchanging color channels. It can be seamlessly integrated into operations, consistently improving robustness against variations. For cross-modality metric learning, we design an...

Person re-identification (ReID) with descriptive query (text or sketch) provides an important supplement for general image-image paradigms, which is usually studied in a single cross-modality matching manner, e.g., text-to-image sketch-to-photo. However, without camera-captured photo query, it uncertain whether the text sketch available not practical scenarios. This motivates us to study new and challenging modality-agnostic person re-ideruification problem. Towards this goal, we propose...

Video-based person re-identification (ReID) matches the same people across video sequences with rich spatial and temporal information in complex scenes. It is highly challenging to capture discriminative when occlusions pose variations exist between frames. A key solution this problem rests on extracting invariant features of sequences. In paper, we propose a novel method for discovering coherence by designing region-level saliency granularity mining network (SGMN). Firstly, address varying...

Existing Weakly-Supervised Change Detection (WSCD) methods often encounter the problem of "instance lumping" under scene-level supervision, particularly in scenarios with a dense distribution changed instances (i.e., objects). In these scenarios, unchanged pixels between are also mistakenly identified as changed, causing multiple changes to be viewed one. practical applications, this issue prevents accurate quantification number changes. To address issue, we propose Dense Instance Separation...

Matching hand-drawn sketches with photos (a.k.a sketch-photo recognition or re-identification) faces the information asymmetry challenge due to abstract nature of sketch modality. Existing works tend learn shared embedding spaces CNN models by discarding appearance cues for photo images introducing GAN synthesis. The former unavoidably loses discriminability, while latter contains ineffaceable generation noise. In this paper, we start first attempt design an information-aligned transformer...

The RGB-D cross-modal person re-identification (re-id) task aims to identify the of interest across RGB and depth image modes. tremendous discrepancy between these two modalities makes this difficult tackle. Few researchers pay attention task, deep networks existing methods still cannot be trained in an end-to-end manner. Therefore, article proposes module for re-id. This network introduces a relational branch narrow gaps heterogeneous images. It models abundant correlations any sample...

Sketch-photo recognition is a cross-modal matching problem whose query sets are sketch images drawn by artists or amateurs. Due to the significant modality difference between two modalities, it challenging extract discriminative modality-shared feature representations. Existing works focus on exploring modality-invariant features discover shared embedding space. However, they discard modality-specific cues, resulting in information loss and diminished discriminatory power of features. This...

The visible-infrared pedestrian re- identification (VI Re-ID) task aims to match cross-modality images with the same labels. Most current methods focus on mitigating modality discrepancy by adopting a two-stream network and identity supervision. Based methods, we propose novel feature fusion center aggregation learning (<inline-formula> <tex-math notation="LaTeX">$F^{2}$ </tex-math></inline-formula>CALNet) for identification. <inline-formula> </tex-math></inline-formula>CALNet focuses...

Unsupervised visible-infrared person re-identification (US-VI-ReID) aims at learning a cross-modality matching model under unsupervised conditions, which is an extremely important task for practical nighttime surveillance to retrieve specific identity. Previous advanced US-VI-ReID works mainly focus on associating the positive identities optimize feature extractor by off-line manners, inevitably resulting in error accumulation of incorrect associations each training epoch due intra-modality...

Infrared-Visible person Re-IDentification (IV-ReID) is an emerging subject, which has important research significance for nighttime monitoring. Existing works focus on reducing cross-modality discrepancies, but the discrepancy cannot be completely eliminated. Therefore, we concentrate excavating commonalities to handle task. Since similar features between two modalities are possessed of commonalities, our goal find more in infrared and visible images. A novel Dual-stream Multi-layer...

Existing approaches usually form the tracking task as an appearance matching procedure. However, discrimination ability of features is insufficient in these trackers, which caused by their weak feature supervision constraints and inadequate exploitation spatial contexts. To tackle this issue, article proposes a novel (AMT) method to strengthen restraints capture discriminative representations. Specifically, we first utilize triplet structural loss function, improves learning capability...

Person re-identification (re-id) is one of the hottest research topics due to its great value in video analysis applications, such as indoor security and road surveillance. It has been verified beneficial for re-id joint global local features recent literature. However, most existing methods usually extract region or divide whole image into several parts without considering alignment different parts, which are not discriminating robust complex scenarios. In this paper, we propose a novel...