In recent years a vast amount of visual content has been generated and shared from many fields, such as social media platforms, medical imaging, and robotics. This abundance of content creation and sharing has introduced new challenges, particularly that of searching databases for similar content - Content Based Image Retrieval (CBIR) - a long-established research area in which improved efficiency and accuracy are needed for real-time retrieval. Artificial intelligence has made progress in CBIR and has significantly facilitated the process of instance search. In this survey we review recent instance retrieval works that are developed based on deep learning algorithms and techniques, with the survey organized by deep feature extraction, feature embedding and aggregation methods, and network fine-tuning strategies. Our survey considers a wide variety of recent methods, whereby we identify milestone work, reveal connections among various methods and present the commonly used benchmarks, evaluation results, common challenges, and propose promising future directions.

Fine-grained image retrieval has been extensively explored in a zero-shot manner. A deep model is trained on the seen part and then evaluated the generalization performance on the unseen part. However, this setting is infeasible for many real-world applications since (1) the retrieval dataset can be non-fixed so that new data are added constantly, and (2) data samples of the seen categories are also common in practice and are important for evaluation. In this paper, we explore lifelong fine-grained image retrieval (LFGIR), which learns continuously on a sequence of new tasks with data from different datasets. We first use knowledge distillation to minimize catastrophic forgetting on old tasks. Training continuously on different datasets causes large domain shifts between the old and new tasks while image retrieval is sensitive to even small shifts in the features. This tends to weaken the effectiveness of knowledge distillation by the frozen teacher. To mitigate the impact of domain shifts, we use the network inversion method to generate images of the old tasks. In addition, we design an on-the-fly teacher which transfers knowledge captured on a new task to the student to improve better generalization performance, thereby achieving a better balance between old and new tasks in the end. We name the whole framework as Dual Knowledge Distillation (DKD), whose efficacy is demonstrated by extensive experimental results on sequential tasks including seven datasets.

Current VQA models are suffering from the problem of overdependence on language bias, which severely reduces their robustness in real-world scenarios. In this paper, we analyze VQA models from the view of forward/backward chaining in the inference engine, and propose to enhance their robustness via a novel Bidirectional Chaining (VQA-BC) framework. Specifically, we introduce a backward chaining with hardnegative contrastive learning to reason from the consequence (answers) to generate crucial known facts (question-related visual region features). Furthermore, to alleviate the overconfident problem in answer prediction (forward chaining), we present a novel introspective regularization to connect forward and backward chaining with label smoothing. Extensive experiments verify that VQA-BC not only effectively overcomes language bias on out-of-distribution dataset, but also alleviates the over-correct problem caused by ensemble-based method on in-distribution dataset. Compared with competitive debiasing strategies, our method achieves state-of-the-art performance to reduce language bias on VQA-CP v2 dataset.

Lifelong person re-identification (LReID) is a challenging and emerging task, which concerns the ReID capability on both seen and unseen domains after learning across different domains continually. Existing works on LReID are devoted to introducing commonly-used lifelong learning approaches, while neglecting a serious side effect caused by using normalization layers in the context of domain-incremental learning. In this work, we aim to raise awareness of the importance of training proper batch normalization layers by proposing a new meta reconciliation normalization (MRN) method specifically designed for tackling LReID. Our MRN consists of grouped mixture standardization and additive rectified rescaling components, which are able to automatically maintain an optimal balance between domain-dependent and domain-independent statistics, and even adapt MRN for different testing instances. Furthermore, inspired by synaptic plasticity in human brain, we present a MRN-based meta-learning framework for mining the meta-knowledge shared across different domains, even without replaying any previous data, and further improve the model's LReID ability with theoretical analyses. Our method achieves new state-of-the-art performances on both balanced and imbalanced LReID benchmarks.

Deep learning for fine-grained image retrieval in an incremental context is less investigated. In this paper, we explore this task to realize the model’s continuous retrieval ability. That means, the model enables to perform well on new incoming data and reduce forgetting of the knowledge learned on preceding old tasks. For this purpose, we distill semantic correlations knowledge among the representations extracted from the new data only so as to regularize the parameters updates using the teacher-student framework. In particular, for the case of learning multiple tasks sequentially, aside from the correlations distilled from the penultimate model, we estimate the representations for all prior models and further their semantic correlations by using the representations extracted from the new data. To this end, the estimated correlations are used as an additional regularization and further prevent catastrophic forgetting over all previous tasks, and it is unnecessary to save the stream of models trained on these tasks. Extensive experiments demonstrate that the proposed method performs favorably for retaining performance on the already-trained old tasks and achieving good accuracy on the current task when new data are added at once or sequentially.

Person re-identification (ReID) methods always learn through a stationary domain that is fixed by the choice of a given dataset. In many contexts (e.g., lifelong learning), those methods are ineffective because the domain is continually changing in which case incremental learning over multiple domains is required potentially. In this work we explore a new and challenging ReID task, namely lifelong person re-identification (LReID), which enables to learn continuously across multiple domains and even generalise on new and unseen domains. Following the cognitive processes in the human brain, we design an Adaptive Knowledge Accumulation (AKA) framework that is endowed with two crucial abilities: knowledge representation and knowledge operation. Our method alleviates catastrophic forgetting on seen domains and demonstrates the ability to generalize to unseen domains. Correspondingly, we also provide a new and large-scale benchmark for LReID. Extensive experiments demonstrate our method outperforms other competitors by a margin of 5.8% mAP in generalising evaluation. The codes will be available at https://github.com/TPCD/LifelongReID.

Most Visual Question Answering (VQA) models are faced with language bias when learning to answer a given question, thereby failing to understand multimodal knowledge simultaneously. Based on the fact that VQA samples with different levels of language bias contribute differently for answer prediction, in this paper, we overcome the language prior problem by proposing a novel Language Bias driven Curriculum Learning (LBCL) approach, which employs an easy-to-hard learning strategy with a novel difficulty metric Visual Sensitive Coefficient (VSC). Specifically, in the initial training stage, the VQA model mainly learns the superficial textual correlations between questions and answers (easy concept) from more-biased examples, and then progressively focuses on learning the multimodal reasoning (hard concept) from less-biased examples in the following stages. The curriculum selection of examples on different stages is according to our proposed difficulty metric VSC, which is to evaluate the difficulty driven by the language bias of each VQA sample. Furthermore, to avoid the catastrophic forgetting of the learned concept during the multi-stage learning procedure, we propose to integrate knowledge distillation into the curriculum learning framework. Extensive experiments show that our LBCL can be generally applied to common VQA baseline models, and achieves remarkably better performance on the VQA-CP v1 and v2 datasets, with an overall 20% accuracy boost over baseline models.

The focus of this survey is on the analysis of two modalities of multimodal deep learning: image and text. Unlike classic reviews of deep learning where monomodal image classifiers such as VGG, ResNet and Inception module are central topics, this paper will examine recent multimodal deep models and structures, including auto-encoders, generative adversarial nets and their variants. These models go beyond the simple image classifiers in which they can do uni-directional (e.g. image captioning, image generation) and bi-directional (e.g. cross-modal retrieval, visual question answering) multimodal tasks. Besides, we analyze two aspects of the challenge in terms of better content understanding in deep multimodal applications. We then introduce current ideas and trends in deep multimodal feature learning, such as feature embedding approaches and objective function design, which are crucial in overcoming the aforementioned challenges. Finally, we include several promising directions for future research.

Accurately matching visual and textual data in cross-modal retrieval has been widely studied in the multimedia community. To address these challenges posited by the heterogeneity gap and the semantic gap, we propose integrating Shannon information theory and adversarial learning. In terms of the heterogeneity gap, we integrate modality classification and information entropy maximization adversarially. For this purpose, a modality classifier (as a discriminator) is built to distinguish the text and image modalities according to their different statistical properties. This discriminator uses its output probabilities to compute Shannon information entropy, which measures the uncertainty of the modality classification it performs. Moreover, feature encoders (as a generator) project uni-modal features into a commonly shared space and attempt to fool the discriminator by maximizing its output information entropy. Thus, maximizing information entropy gradually reduces the distribution discrepancy of cross-modal features, thereby achieving a domain confusion state where the discriminator cannot classify two modalities confidently. To reduce the semantic gap, Kullback-Leibler (KL) divergence and bi-directional triplet loss are used to associate the intra- and inter-modality similarity between features in the shared space. Furthermore, a regularization term based on KL-divergence with temperature scaling is used to calibrate the biased label classifier caused by the data imbalance issue. Extensive experiments with four deep models on four benchmarks are conducted to demonstrate the effectiveness of the proposed approach.

Convolutional neural network training can suffer from diverse issues like exploding or vanishing gradients, scaling-based weight space symmetry and covariant-shift. In order to address these issues, researchers develop weight regularization methods and activation normalization methods. In this work we propose a weight soft-regularization method based on the Oblique manifold. The proposed method uses a loss function which pushes each weight vector to have a norm close to one, i.e. the weight matrix is smoothly steered toward the so-called Oblique manifold. We evaluate our method on the very popular CIFAR-10, CIFAR-100 and ImageNet 2012 datasets using two state-of-the-art architectures, namely the ResNet and wide-ResNet. Our method introduces negligible computational overhead and the results show that it is competitive to the state-of-the-art and in some cases superior to it. Additionally, the results are less sensitive to hyperparameter settings such as batch size and regularization factor.

Computational Fluid Dynamics (CFD) simulations are a very important tool for many industrial applications, such as aerodynamic optimization of engineering designs like cars shapes, airplanes parts etc. The output of such simulations, in particular the calculated flow fields, are usually very complex and hard to interpret for realistic three-dimensional real-world applications, especially if time-dependent simulations are investigated. Automated data analysis methods are warranted but a non-trivial obstacle is given by the very large dimensionality of the data. A flow field typically consists of six measurement values for each point of the computational grid in 3D space and time (velocity vector values, turbulent kinetic energy, pressure and viscosity). In this paper we address the task of extracting meaningful results in an automated manner from such high dimensional data sets. We propose deep learning methods which are capable of processing such data and which can be trained to solve relevant tasks on simulation data, i.e. predicting drag and lift forces applied on an airfoil. We also propose an adaptation of the classical hand crafted features known from computer vision to address the same problem and compare a large variety of descriptors and detectors. Finally, we compile a large dataset of 2D simulations of the flow field around airfoils which contains 16000 flow fields with which we tested and compared approaches. Our results show that the deep learning-based methods, as well as hand crafted feature based approaches, are well-capable to accurately describe the content of the CFD simulation output on the proposed dataset.

Visible-infrared person re-identification (VI-ReID) is a challenging and essential task in night-time intelligent surveillance systems. Except for the intra-modality variance that RGB-RGB person re-identification mainly overcomes, VI-ReID suffers from additional inter-modality variance caused by the inherent heterogeneous gap. To solve the problem, we present a carefully designed dual Gaussian-based variational auto-encoder (DG-VAE), which disentangles an identity-discriminable and an identity-ambiguous cross-modality feature subspace, following a mixture-of-Gaussians (MoG) prior and a standard Gaussian distribution prior, respectively. Disentangling cross-modality identity-discriminable features leads to more robust retrieval for VI-ReID. To achieve efficient optimization like conventional VAE, we theoretically derive two variational inference terms for the MoG prior under the supervised setting, which not only restricts the identity-discriminable subspace so that the model explicitly handles the cross-modality intra-identity variance, but also enables the MoG distribution to avoid posterior collapse. Furthermore, we propose a triplet swap reconstruction (TSR) strategy to promote the above disentangling process. Extensive experiments demonstrate that our method outperforms state-of-the-art methods on two VI-ReID datasets. Codes will be available at https://github.com/TPCD/DG-VAE.

Cross-modal hash retrieval has received considerable interest in the area of deep learning. Here hash codes of data of different modalities are learned where pair-wise loss functions control feature similarity in a shared embedding space. In this paper we improve on feature similarity by using Shannon's information entropy with respect to the modality information that is left in learning superior hash codes. We introduce a novel network for predicting the domain from the learned features while the protagonist network uses a loss function based on Shannon's information entropy to learn to maximize the domain uncertainty and therefore the information content. Additionally, according to the number of common labels between each similar image-text pair, we define a multi-level similarity matrix as supervisory information, which constrains all similar pairs with different weights. We show with extensive experiments that our novel approach to domain uncertainty leads to a cross-modal hash retrieval that outperforms the state-of-the-art.

Fashion style transfer has attracted significant attention because it both has interesting scientific challenges and it is also important to the fashion industry. This paper focuses on addressing a practical problem in fashion style transfer, person-to-person clothing swapping, which aims to visualize what the person would look like with the target clothes worn on another person instead of dressing them physically. This problem remains challenging due to varying pose deformations between different person images. In contrast to traditional nonparametric methods that blend or warp the target clothes for the reference person, in this paper we propose a multistage deep generative approach named SwapGAN that exploits three generators and one discriminator in a unified framework to fulfill the task end-to-end. The first and second generators are conditioned on a human pose map and a segmentation map, respectively, so that we can simultaneously transfer the pose style and the clothes style. In addition, the third generator is used to preserve the human body shape during the image synthesis process. The discriminator needs to distinguish two fake image pairs from the real image pair. The entire SwapGAN is trained by integrating the adversarial loss and the mask-consistency loss. The experimental results on the DeepFashion dataset demonstrate the improvements of SwapGAN over other existing approaches through both quantitative and qualitative evaluations. Moreover, we conduct ablation studies on SwapGAN and provide a detailed analysis about its effectiveness.

Image-to-image translation between different domains has been an important research direction, with the aim of arbitrarily manipulating the source image content to become similar to a target image. Recently, cycle-consistent generative network (CycleGAN) has become a fundamental approach for general-purpose image-to-image translation, while almost no work has examined what factors may influence its performance. To provide more insights, we propose two new models roughly based on CycleGAN, namely Long CycleGAN and Nest CycleGAN. First, Long CycleGAN cascades several generators to perform the domain translation in a long cycle. It shows the benefit of stacking more generators on the generation quality. In addition to the long cycle, Nest CycleGAN develops new inner cycles to bridge intermediate generators directly, which can help constrain the unsupervised mappings. In the experiments, we conduct qualitative and quantitative comparisons for tasks including photo↔label, photo↔sketch, and photo colorization. The quantitative and qualitative results demonstrate the effectiveness of our two proposed models.