Object counting is a fundamental task in computer vision, with critical applications in areas such as crowd monitoring and ecological conservation. Traditional methods typically rely on large-scale annotated datasets, which are costly and time-consuming to obtain. Few-shot object counting has emerged as a promising solution, enabling accurate counting with minimal annotated samples. However, in real-world scenarios, objects often exhibit significant scale variations due to factors such as view distortion, varying shooting distances, and inherent size differences. Existing few-shot methods usually struggle to address this challenge effectively. To address these issues, we propose a Scale-Aware Vision Transformer (SAViT) framework. Specifically, we design a multi-scale dilated convolution module in SAViT, which can adaptively adjust convolution kernel sampling rates to handle objects of varying sizes. Additionally, we incorporate a global channel attention mechanism to strengthen the model’s ability to capture robust feature representations, thereby improving detection accuracy. For practical usability, we integrate the Segment Anything Model (SAM) to create an exemplar box selection module, simplifying the process by allowing users to generate precise exemplar boxes with a single line drawn on the target object. Extensive experiments on the FSC-147 dataset demonstrate the effectiveness of our approach, achieving a Mean Absolute Error (MAE) of 8.92 and a Root Mean Squared Error (RMSE) of 31.26. Compared to the state-of-the-art method, CACViT, our model reduces MAE by 0.21 (2.30% improvement) and RMSE by 17.7 (36.15% improvement). Our approach not only provides an effective solution for few-shot object counting but also provides a new practical paradigm for extending few-shot learning to complex vision tasks requiring multi-scale reasoning. The code of our paper is available at https://github.com/BlouseDong/SAViT .

Facial image based kinship verification aims to decide whether there exists kinship between the given facial images. In practice, the cross-generation differences will cause adverse effects on kinship verification, which limits the performance. Therefore, how to mine the implied similarity from facial images with large cross-generation divergence is an important problem in kinship verification, which has not yet been well studied. In view of this, we propose a Similarity Mining via Implicit matching pattern LEarning (SMILE) approach for kinship verification. Specifically, SMILE mainly consists of two modules, including a Semi-coupled Multi-pattern Similarity Learning (SMSL) module and a Cross-Generation Feature Normalization (CGFN) module. The SMSL module is designed to learn multiple semi-coupled matching patterns for mining the implicit facial similarity information from different perspectives. The CGFN module aims to reduce the divergence between facial images of parent and child. Extensive experiments demonstrate that the proposed approach outperforms the existing state-of-the-art methods.

Cloth-Changing Person Re-Identification (CC-ReID) refers to the technology that can identify the same individual in surveillance scenarios, even after changing their clothes. The core of CC-ReID is to exact cloth-irrelevant identity features. Existing methods usually extract them from body contours, skeletons, human faces, etc. However, information extracted from a single modality is often insufficient. Therefore, we propose a novel Multimodal Interaction Fusion (MIF) framework that could facilitate the deep integration of multimodal information, thereby enhancing the generalization ability of CC-ReID. Specifically, we design a triple-interaction collaborative module that strategically integrates features captured from body posture, original RGB images, and clothes-erased images, maximizing the advantages of each modality and overcoming their limitations. Additionally, we develop an Adapt-Blend Pool (ABP) module that effectively improves pooling methods to enhance the model’s ability to process images with complex backgrounds and varied features. Following training, our framework effectively harnesses a variety of modal inputs to capture identity representations independent of clothing accurately. Extensive experimental results on PRCC, LTCC, and VC-Clothes datasets demonstrate the effectiveness of the proposed method.

Protein-protein interaction (PPI) plays an important role in almost all life activities. Many protein interaction sites have been confirmed by biological experiments, but these PPI site identification methods are time-consuming and expensive. In this study, a deep learning-based PPI prediction method, named DeepSG2PPI, is developed. First, the protein sequence information is retrieved and the local context information of each amino acid residue is calculated. A two-dimensional convolutional neural network (2D-CNN) model is employed to extract features from a two-channel coding structure, in which an attention mechanism is embedded to assign higher weights to key features. Second, the global statistical information of each amino acid residue and the relationship graph between the protein and GO (Gene Ontology) function annotation are built, and the graph embedding vector is constructed to represent the biological features of the protein. Finally, a 2D-CNN model and two 1D-CNN models are combined for PPI prediction. The comparison analysis with existing algorithms shows that the DeepSG2PPI method has better performance. It provides more accurate and effective PPI site prediction, which will be helpful in reducing the cost and failure rate of biological experiments.

Image-based kinship verification is an important task in computer vision and has many applications in practice, such as missing children search and family album construction, among others. Due to the differences in age, gender, expression and appearance, there usually exists a large discrepancy between the facial images of parent and child. This makes kinship verification a challenging task. In this article, we propose a Distance and Direction Based Deep Discriminant Metric Learning (D 4 ML) approach for kinship verification. The basic idea of D 4 ML is to make full use of the discriminant information contained in the facial images of parent and child such that the network can learn more a discriminating distance metric. Specifically, D 4 ML learns the metric by utilizing the discriminant information from two perspectives: distance-based perspective and direction-based perspective. From the distance-based perspective, the designed loss function is used to minimize the distance between images having kinship and maximize the distance between images without kinship. In practice, the gender difference and large age gap may significantly increase the distance between facial images of parent and child. Therefore, learning the metric only from a distance-based perspective is insufficient. Considering that two vectors with a large distance may appear with high similarity in direction, D 4 ML also employs the direction-based loss function in the training process. Both kinds of loss function work together to improve the discriminability of the learned metric. Experimental results on four small size publicly available datasets demonstrate the effectiveness of our approach. Source code of our approach can be found at https://github.com/lclhenu/D4ML .

Kinship verification is an interesting and important problem in the fields of computer vision. In practice, the biggest obstacle in kinship verification is that the representation capability of extracted features may not be powerful due to the significant differences between facial images of family members. To effectively address this problem, we propose a semi-coupled synthesis and analysis dictionary pair learning (SSADL) approach, which can reduce the differences between facial images. Specifically, SSADL jointly learns two view-specific synthesis-analysis dictionary pairs as well as a mapping matrix from the training data of parent and child, with which, the heterogeneous facial images of parent and child can be transformed into coding coefficients of the same subspace, such that the kinship verification task can be conducted using the coding coefficients. Besides, we also design a hard sample based coefficient discriminant term to ensure that the obtained coefficients own favorable discriminability. Experimental results on several publicly used benchmarks show the effectiveness of our proposed approach.