Convolutional neural networks (CNNs) have achieved significant performance on various real-life tasks. However, the large number of parameters in convolutional layers requires huge storage and computation resources, making it challenging to deploy CNNs on memory-constrained embedded devices. In this article, we propose a novel compression method that generates the convolution filters in each layer using a set of learnable low-dimensional quantized filter bases. The proposed method reconstructs the convolution filters by stacking the linear combinations of these filter bases. By using quantized values in weights, the compact filters can be represented using fewer bits so that the network can be highly compressed. Furthermore, we explore the sparsity of coefficients through $L_{1}$ -ball projection when conducting linear combination to further reduce the storage consumption and prevent overfitting. We also provide a detailed analysis of the compression performance of the proposed method. Evaluations of image classification and object detection tasks using various network structures demonstrate that the proposed method achieves a higher compression ratio with comparable accuracy compared with the existing state-of-the-art filter decomposition and network quantization methods.

The imbalanced distribution of long-tailed data presents a considerable challenge for deep learning models, as it causes them to prioritize the accurate classification of head classes but largely disregard tail classes. The biased decision boundary caused by inadequate semantic information in tail classes is one of the key factors contributing to their low recognition accuracy. To rectify this issue, we propose to augment tail classes by grafting the diverse semantic information from head classes, referred to as head-to-tail fusion (H2T). We replace a portion of feature maps from tail classes with those belonging to head classes. These fused features substantially enhance the diversity of tail classes. Both theoretical analysis and practical experimentation demonstrate that H2T can contribute to a more optimized solution for the decision boundary. We seamlessly integrate H2T in the classifier adjustment stage, making it a plug-and-play module. Its simplicity and ease of implementation allow for smooth integration with existing long-tailed recognition methods, facilitating a further performance boost. Extensive experiments on various long-tailed benchmarks demonstrate the effectiveness of the proposed H2T. The source code is available at https://github.com/Keke921/H2T.

Supervised cross-modal hashing methods usually construct a massive undirected weighted graph based on labels for training data, with the aim of learning more structured hash codes by preserving relationships within this graph. However, as the volume of data increases, such an approach demands substantial computational and storage resources and tends to aggregate all data points with paths, even semantically unrelated ones, which undermines the retrieval performance. In this paper, we propose to prune less crucial paths from this graph to obtain a clearer representation of relationships among data points. This not only reduces computational resources but separates semantically unrelated data points. Specifically, we define key points within the graph and retain relationships only between all data points and these key points, resulting in a simplified and more transparent graph that is used to supervise hash code learning. Experimental results on three datasets demonstrate that removing unimportant paths from the relationship graph can lead to the learning of more structured hash codes, thereby improving retrieval performance.

Cross-modal hashing (CMH) has attracted considerable attention in recent years. Almost all existing CMH methods primarily focus on reducing the modality gap and semantic gap, i.e., aligning multi-modal features and their semantics in Hamming space, without taking into account the space gap, i.e., difference between the real number space and the Hamming space. In fact, the space gap can affect the performance of CMH methods. In this paper, we analyze and demonstrate how the space gap affects the existing CMH methods, which therefore raises two problems: solution space compression and loss function oscillation. These two problems eventually cause the retrieval performance deteriorating. Based on these findings, we propose a novel algorithm, namely Semantic Channel Hashing (SCH). First, we classify sample pairs into fully semantic-similar, partially semantic-similar, and semantic-negative ones based on their similarity and impose different constraints on them, respectively, to ensure that the entire Hamming space is utilized. Then, we introduce a semantic channel to alleviate the issue of loss function oscillation. Experimental results on three public datasets demonstrate that SCH outperforms the state-of-the-art methods. Furthermore, experimental validations are provided to substantiate the conjectures regarding solution space compression and loss function oscillation, offering visual evidence of their impact on the CMH methods.

As an effective tool for network compression, pruning techniques have been widely used to reduce the large number of parameters in deep neural networks (NNs). Nevertheless, unstructured pruning has the limitation of dealing with the sparse and irregular weights. By contrast, structured pruning can help eliminate this drawback but it requires complex criteria to determine which components to be pruned. Therefore, this paper presents a new method termed BUnit-Net, which directly constructs compact NNs by stacking designed basic units, without requiring additional judgement criteria anymore. Given the basic units of various architectures, they are combined and stacked systematically to build up compact NNs which involve fewer weight parameters due to the independence among the units. In this way, BUnit-Net can achieve the same compression effect as unstructured pruning while the weight tensors can still remain regular and dense. We formulate BUnit-Net in diverse popular backbones in comparison with the state-of-the-art pruning methods on different benchmark datasets. Moreover, two new metrics are proposed to evaluate the trade-off of compression performance. Experiment results show that BUnit-Net can achieve comparable classification accuracy while saving around 80% FLOPs and 73% parameters. That is, stacking basic units provides a new promising way for network compression.

Supervised cross-modal hashing has received wide attention in recent years. However, existing methods primarily rely on sample-wise semantic relationships to evaluate the semantic similarity between samples, overlooking the impact of label distribution on enhancing retrieval performance. Moreover, the limited representation capability of traditional dense hash codes hinders the preservation of semantic relationship. To overcome these challenges, we propose a new method, Joint Semantic Preserving Sparse Hashing (JSPSH). Specifically, we introduce a new concept of cluster-wise semantic relationship, which leverages label distribution to indicate which samples are more suitable for clustering. Then, we jointly utilize sample-wise and cluster-wise semantic relationships to supervise the learning of hash codes. In this way, JSPSH preserves both kinds of semantic relationships to ensure that more samples with similar semantics are clustered together, thereby achieving better retrieval results. Furthermore, we utilize high-dimensional sparse hash codes that offer stronger representation capability to preserve such more complex semantics. Finally, an interaction term is introduced in hash functions learning stage to further narrow the gap between modalities. Experimental results on three large-scale datasets demonstrate the effectiveness of JSPSH in achieving superior retrieval performance. Codes are available at https://github.com/hutt94/JSPSH.

Lip-password has provided a promising solution for speaker verification (Liu and Cheung 2014). Despite the potential of this technology, there are few related studies, largely attributed to the lack of corresponding public datasets. Furthermore, previous works in this field generally demand a substantial amount of training samples and negative samples, impeding their applications from a practical perspective. Therefore, this paper collects a lip-password dataset and proposes a novel few-shot lip-password based speaker verification model, which can be effectively deployed in real-world scenarios because only a small number of data are required for training. Specifically, with an analysis of lip-password features, a down-sampling strategy is presented to generate more training samples. To compensate for the information loss caused by this strategy, a few-shot model, consisting of global and local models, is designed to simultaneously verify the global and local information of the lip-password. Speaker identity is verified only if both stages are passed. The efficacy of the proposed method is demonstrated using the newly collected dataset.

Deep Convolutional Neural Networks (CNN) have been successfully applied to many real-life problems. However, the huge memory cost of deep CNN models poses a great challenge of deploying them on memory-constrained devices (e.g., mobile phones). One popular way to reduce the memory cost of deep CNN model is to train binary CNN where the weights in convolution filters are either 1 or -1 and therefore each weight can be efficiently stored using a single bit. However, the compression ratio of existing binary CNN models is upper bounded by ∼ 32. To address this limitation, we propose a novel method to compress deep CNN model by stacking low-dimensional binary convolution filters. Our proposed method approximates a standard convolution filter by selecting and stacking filters from a set of low-dimensional binary convolution filters. This set of low-dimensional binary convolution filters is shared across all filters for a given convolution layer. Therefore, our method will achieve much larger compression ratio than binary CNN models. In order to train our proposed model, we have theoretically shown that our proposed model is equivalent to select and stack intermediate feature maps generated by low-dimensional binary filters. Therefore, our proposed model can be efficiently trained using the split-transform-merge strategy. We also provide detailed analysis of the memory and computation cost of our model in model inference. We compared the proposed method with other five popular model compression techniques on two benchmark datasets. Our experimental results have demonstrated that our proposed method achieves much higher compression ratio than existing methods while maintains comparable accuracy.