Salient Object Detection (SOD) has been widely used in practical applications such as multi-sensor image fusion, remote sensing, and defect detection. Recently, SOD from RGB Thermal (T) rapidly developed due to its robustness extreme situations like low illumination occlusion. However, existing methods all utilize a dual-stream encoder, which significantly increases the computation burdens hinders real-world deployment. To this end, we propose real-time One-stream Semantic-guided Refinement...

RGB-T salient object detection (SOD) aims at utilizing the complementary cues of RGB and Thermal (T) modalities to detect segment common objects. However, on one hand, existing methods simply fuse features two without fully considering characters T. On other high computational cost prevents them from real-world applications (e.g., automatic driving, abnormal detection, person re-ID). To this end, we proposed an efficient encoder-decoder network named Context-guided Stacked Refinement Network...

Raw underwater images suffer from low contrast and color cast due to wavelength-selective light scattering attenuation. The distortions in luminance mainly appear at the frequency while that edge texture are high frequency. However, hybrid difficult simultaneously recover for existing methods, which focus on spatial domain. To tackle these issues, we propose a novel deep learning network progressively refine by wavelet boost strategy (PRWNet), both domains. Specifically, Multi-stage...

Haze can seriously affect the visible and visual quality of outdoor optical sensor systems, e.g., driving assistance, remote sensing, video surveillance. Single image dehazing is an intractable problem due to its ill-posed nature. The main idea paper combining multi-scale fusion strategy prior knowledge, thereby presenting balanced contrast enhancement intrinsic color preservation, efficiently. atmospheric illumination (AIP) has been proved that haze mainly degrades luminance channel rather...

Zero-shot learning (ZSL) is an extreme case of transfer that aims to recognize samples (e.g., images) unseen classes relying on a train-set covering only seen and set auxiliary knowledge semantic descriptors). Existing methods usually resort constructing visual-to-semantics mapping based features extracted from each whole sample. However, since the visual spaces are inherently independent may exist in different manifolds, these easily suffer domain bias problem due classes. Unlike existing...

This paper investigates a new, practical, but challenging problem named Non-exemplar Online Class-incremental continual Learning (NO-CL), which aims to preserve the discernibility of base classes without buffering data examples and efficiently learn novel continuously in single-pass (i.e., online) stream. The challenges this task are mainly two-fold: (1) Both suffer from severe catastrophic forgetting as no previous samples available for replay. (2) As online can only be observed once, there...

Compositional Zero-shot Learning (CZSL) aims to recognize novel concepts composed of known knowledge without training samples. Standard CZSL either identifies visual primitives or enhances unseen entities, and as a result, entanglement between state object cannot be fully utilized. Admittedly, vision-language models (VLMs) could naturally cope with through tuning prompts, while uneven leads prompts dragged into local optimum. In this paper, we take further step introduce Disentangled...

Multi-label zero-shot learning extends conventional single-label to a more realistic scenario that aims at recognizing multiple unseen labels of classes for each input sample. Existing works usually exploit attention mechanism generate the correlation among different labels. However, most them are biased on several <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">major classes</i> while neglect xmlns:xlink="http://www.w3.org/1999/xlink">minor...

Blind image quality assessment suffers from the range effect, which indicates that on overall range, mean opinion score and predicted MOS are well correlated while focusing a particular narrow correlation is lower. To tackle this problem, novel method proposed coarse-grained metric to fine-grained prediction. Concretely, we utilize global context features local detailed for multi-scale distortion perception. Then, further boost assessment, introduce feedback mechanism, in accord with Human...

Recent studies usually approach multi-label zeroshot learning (MLZSL) with visual-semantic mapping on spatial-class correlation, which can be computationally costly, and worse still, fails to capture fine-grained classspecific semantics. We observe that different channels may have sensitivities classes, correspond specific Such an intrinsic channelclass correlation suggests a potential alternative for the more accurate class-harmonious feature representations. In this paper, our interest is...

Unexploded ordnance (UXO) survey is the foremost task of clearance project. Transient electromagnetic method (TEM) proved effective for UXO survey. However, it still difficult TEM to detect small-size, ultra-shallow and dense targets because large-size devices inversion methods large-scale applications are usually ineffective. In work, in order avoid complex apparent resistivity, voltages acquired by our specified small-loop system used depict a voltage distribution profile, which then...

Open-World Compositional Zero-shot Learning (OW-CZSL) aims to recognize novel compositions of state and object primitives in images with no priors on the compositional space, which induces a tremendously large output space containing all possible state-object compositions. Existing works either learn joint embedding or predict simple separate classifiers. However, former method heavily relies external word methods, latter ignores interactions interdependent primitives, respectively. In this...

While Large Vision-Language Models (LVLMs) have rapidly advanced in recent years, the prevalent issue known as `hallucination' problem has emerged a significant bottleneck, hindering their real-world deployments. Existing methods mitigate this mainly from two perspectives: One approach leverages extra knowledge like robust instruction tuning LVLMs with curated datasets or employing auxiliary analysis networks, which inevitable incur additional costs. Another approach, contrastive decoding,...

This paper investigates a new, practical, but challenging problem named Non-exemplar Online Class-incremental continual Learning (NO-CL), which aims to preserve the discernibility of base classes without buffering data examples and efficiently learn novel continuously in single-pass (i.e., online) stream. The challenges this task are mainly two-fold: (1) Both suffer from severe catastrophic forgetting as no previous samples available for replay. (2) As online can only be observed once, there...

Multi-label zero-shot learning extends conventional single-label to a more realistic scenario that aims at recognizing multiple unseen labels of classes for each input sample. Existing works usually exploit attention mechanism generate the correlation among different labels. However, most them are biased on several major while neglect minor with same importance in samples, and may thus result overly diffused maps cannot sufficiently cover classes. We argue disregarding connection between...

Most existing trackers based on correlation filtering try to introduce different regularization items improve the learning process of tracking object. However, parameters corresponding need be adjusted dynamically meet process. The are often fixed, so they cannot updated for more accurate tracking. We propose a method adaptively adjust spatiotemporal during make use important between response map and quality in confidence description index optimize parameters, training can focus part with...

Blind image quality assessment (BIQA) of user generated content (UGC) suffers from the range effect which indicates that on overall range, mean opinion score (MOS) and predicted MOS (pMOS) are well correlated; focusing a particular correlation is lower. The reason for deviations both in wide narrow destroy uniformity between pMOS. To tackle this problem, novel method proposed coarse-grained metric to fine-grained prediction. Firstly, we design rank-and-gradient loss metric. keeps order grad...

Open-World Compositional Zero-shot Learning (OW-CZSL) aims to recognize novel compositions of state and object primitives in images with no priors on the compositional space, which induces a tremendously large output space containing all possible state-object compositions. Existing works either learn joint embedding or predict simple separate classifiers. However, former heavily relies external word methods, latter ignores interactions interdependent primitives, respectively. In this paper,...