Due to the impressive zero-shot capabilities, pre-trained vision-language models (e.g., CLIP), have attracted widespread attention and adoption across various domains. Nonetheless, CLIP has been observed to be susceptible to adversarial examples. Through experimental analysis, we have observed a phenomenon wherein adversarial perturbations induce shifts in text-guided attention. Building upon this observation, we propose a simple yet effective strategy: Text-Guided Attention for Zero-Shot Robustness (TGA-ZSR). This framework incorporates two components: Local Attention Refinement Module and Global Attention Constraint Module. Our goal is to maintain the generalization of the CLIP model and enhance its adversarial robustness: The Local Attention Refinement Module aligns the text-guided attention obtained from the target model via adversarial examples with the text-guided attention acquired from the original model via clean examples. This alignment enhances the model's robustness. Additionally, the Global Attention Constraint Module acquires text-guided attention from both the target and original models using clean examples. Its objective is to maintain model performance on clean samples while enhancing overall robustness. However, we observe that the method occasionally focuses on irrelevant or spurious features, which can lead to suboptimal performance and undermine its robustness in certain scenarios. To overcome this limitation, we further propose a novel approach called Complementary Text-Guided Attention (Comp-TGA). This method integrates two types of foreground attention: attention guided by the class prompt and reversed attention driven by the non-class prompt. These complementary attention mechanisms allow the model to capture a more comprehensive and accurate representation of the foreground. The experiments validate that TGA-ZSR and Comp-TGA yield 9.58% and 11.95% improvements respectively, in zero-shot robust accuracy over the current state-of-the-art techniques across 16 datasets.

Continual learning (CL) aims to enable learning systems to acquire new knowledge constantly without forgetting previously learned information. CL faces the challenge of mitigating catastrophic forgetting while maintaining interpretability across tasks. Most existing CL methods focus primarily on preserving learned knowledge to improve model performance. However, as new information is introduced, the interpretability of the learning process becomes crucial for understanding the evolving decision-making process, yet it is rarely explored. In this paper, we introduce a novel framework that integrates language-guided Concept Bottleneck Models (CBMs) to address both challenges. Our approach leverages the Concept Bottleneck Layer, aligning semantic consistency with CLIP models to learn human-understandable concepts that can generalize across tasks. By focusing on interpretable concepts, our method not only enhances the model’s ability to retain knowledge over time but also provides transparent decision-making insights. We demonstrate the effectiveness of our approach by achieving superior performance on several datasets, outperforming state-of-the-art methods with an improvement of up to 3.06% in final average accuracy on ImageNet-subset. Additionally, we offer concept visualizations for model predictions, further advancing the understanding of interpretable continual learning. Code is available at https://github.com/FisherCats/CLG-CBM.

Online incremental learning aims to enable learning systems to continuously accumulate new knowledge from streaming data in a single-pass manner while preserving previously acquired information. This more realistic and challenging setting has gained increasing attention in recent years. The state-of-the-art methods treat each module of a model, from shallow to deep, as a separate sub-expert network and transfer all the shallow expert knowledge into the final deep expert network. Although this yields notable improvements, we argue that directly supervising shallow layers hampers their acquisition of task-invariant knowledge. Furthermore, explicitly designating the final expert as a student network to absorb knowledge from other experts lacks adaptability, considering that different experts may not excel uniformly across all tasks. To address the aforementioned limitations, we leverage the shallow layers of the model as a shared feature extractor, while the deeper layers form a set of experts capable of learning robust and diverse features. Moreover, to facilitate knowledge transfer between multiple experts, we introduce the LEEP score to assess the feature transferability of each expert on new tasks, thereby selecting the most suitable expert as the teacher network for the new task. Extensive experiments on two evaluation benchmarks verify the effectiveness of our method (e.g, up to 1.3% on Split CIFAR-100 and 2.5% on Split Tiny-ImageNet). Code is available at https://github.com/untitledunmastered1998/MDE-OIL.

Diffusion-based re-ranking methods are effective in modeling the data manifolds through similarity propagation in affinity graphs. However, positive signals tend to diminish over several steps away from the source, reducing discriminative power beyond local regions. To address this issue, we introduce the Locality Preserving Markovian Transition (LPMT) framework, which employs a long-term thermodynamic transition process with multiple states for accurate manifold distance measurement. The proposed LPMT first integrates diffusion processes across separate graphs using Bidirectional Collaborative Diffusion (BCD) to establish strong similarity relationships. Afterwards, Locality State Embedding (LSE) encodes each instance into a distribution for enhanced local consistency. These distributions are interconnected via the Thermodynamic Markovian Transition (TMT) process, enabling efficient global retrieval while maintaining local effectiveness. Experimental results across diverse tasks confirm the effectiveness of LPMT for instance retrieval.

Non-exemplar class incremental learning aims to learn both the new and old tasks without accessing any training data from the past. This strict restriction enlarges the difficulty of alleviating catastrophic forgetting since all techniques can only be applied to current task data. Considering this challenge, we propose a novel framework of fine-grained knowledge selection and restoration. The conventional knowledge distillation-based methods place too strict constraints on the network parameters and features to prevent forgetting, which limits the training of new tasks. To loose this constraint, we proposed a novel fine-grained selective patch-level distillation to adaptively balance plasticity and stability. Some task-agnostic patches can be used to preserve the decision boundary of the old task. While some patches containing the important foreground are favorable for learning the new task. Moreover, we employ a task-agnostic mechanism to generate more realistic prototypes of old tasks with the current task sample for reducing classifier bias for fine-grained knowledge restoration. Extensive experiments on CIFAR100, TinyImageNet and ImageNet-Subset demonstrate the effectiveness of our method. Code is available at https://github.com/scok30/vit-cil.

Due to the impressive zero-shot capabilities, pre-trained vision-language models (e.g. CLIP), have attracted widespread attention and adoption across various domains. Nonetheless, CLIP has been observed to be susceptible to adversarial examples. Through experimental analysis, we have observed a phenomenon wherein adversarial perturbations induce shifts in text-guided attention. Building upon this observation, we propose a simple yet effective strategy: Text-Guided Attention for Zero-Shot Robustness (TGA-ZSR). This framework incorporates two components: the Attention Refinement module and the Attention-based Model Constraint module. Our goal is to maintain the generalization of the CLIP model and enhance its adversarial robustness: The Attention Refinement module aligns the text-guided attention obtained from the target model via adversarial examples with the text-guided attention acquired from the original model via clean examples. This alignment enhances the model’s robustness. Additionally, the Attention-based Model Constraint module acquires text-guided attention from both the target and original models using clean examples. Its objective is to maintain model performance on clean samples while enhancing overall robustness. The experiments validate that our method yields a 9.58% enhancement in zero-shot robust accuracy over the current state-of-the-art techniques across 16 datasets. Our code is available at https://github.com/zhyblue424/TGA-ZSR.

Despite deep neural networks (DNNs) show impressive performance across diverse tasks, they suffer from catastrophic forgetting when dealing with continuous data streams. Incremental learning aims to alleviate this phenomenon and enable DNNs to accumulate new knowledge to cope with the ever-changing world. Recently numerous advanced methods have been developed to enhance the incremental learning capabilities of neural networks. However, these methods mainly focus on the large networks, neglecting the unique needs of edged-device applications, which is surprisingly under-investigated in previous literature. In this paper, we propose two strategies for transferring knowledge from large teacher networks to light-weighted networks in class incremental learning. Specifically, in cases where the initial task contains a large number of categories, our static teacher strategy involves transferring knowledge from the teacher to the student network on the initial task to enhance the plasticity of the student network, and applying regularization constraints on the subsequent task to improve its stability. In a more challenging scenario where each task includes an equal number of categories, the dynamic teacher strategy continuously guides the student network on each task. We evaluate the proposed methods on CIFAR100, Tiny-ImageNet and ImageNet-subset datasets with different types of light-weighted networks (MobileNet, ShuffleNet). We observed that effective knowledge transfer resulting in the student network achieving performance comparable or even outperform the teacher network. Extensive and detailed experiments conducted on three datasets demonstrated the simplicity and effectiveness of our proposed method. Comprehensive analysis are also conducted including different factors and visualization.

Object Re-identification (ReID) is a task focused on retrieving a probe object from a multitude of gallery images using a ReID model trained on a stationary, camera-free dataset. This training involves associating and aggregating identities across various camera views. However, when deploying ReID algorithms in real-world scenarios, several challenges, such as storage constraints, privacy considerations, and dynamic changes in camera setups, can hinder their generalizability and practicality. To address these challenges, we introduce a novel ReID task called Camera-Incremental Object Re-identification (CIOR). In CIOR, we treat each camera's data as a separate source and continually optimize the ReID model as new data streams come from various cameras. By associating and consolidating the knowledge of common identities, our aim is to enhance discrimination capabilities and mitigate the problem of catastrophic forgetting. Therefore, we propose a novel Identity Knowledge Evolution (IKE) framework for CIOR, consisting of Identity Knowledge Association (IKA), Identity Knowledge Distillation (IKD), and Identity Knowledge Update (IKU). IKA is proposed to discover common identities between the current identity and historical identities, facilitating the integration of previously acquired knowledge. IKD involves distilling historical identity knowledge from common identities, enabling rapid adaptation of the historical model to the current camera view. After each camera has been trained, IKU is applied to continually expand identity knowledge by combining historical and current identity memories. Market-CL and Veri-CL evaluations show the effectiveness of Identity Knowledge Evolution (IKE) for CIOR.Code:https://github.com/htyao89/Camera-Incremental-Object-ReID

In class-incremental semantic segmentation, we have no access to the labeled data of previous tasks. Therefore, when incrementally learning new classes, deep neural networks suffer from catastrophic forgetting of previously learned knowledge. To address this problem, we propose to apply a self-training approach that leverages unlabeled data, which is used for rehearsal of previous knowledge. Specifically, we first learn a temporary model for the current task, and then, pseudo labels for the unlabeled data are computed by fusing information from the old model of the previous task and the current temporary model. In addition, conflict reduction is proposed to resolve the conflicts of pseudo labels generated from both the old and temporary models. We show that maximizing self-entropy can further improve results by smoothing the overconfident predictions. Interestingly, in the experiments, we show that the auxiliary data can be different from the training data and that even general-purpose, but diverse auxiliary data can lead to large performance gains. The experiments demonstrate the state-of-the-art results: obtaining a relative gain of up to 114% on Pascal-VOC 2012 and 8.5% on the more challenging ADE20K compared to previous state-of-the-art methods.

Class-incremental learning of deep networks sequentially increases the number of classes to be classified. During training, the network has only access to data of one task at a time, where each task contains several classes. In this setting, networks suffer from catastrophic forgetting which refers to the drastic drop in performance on previous tasks. The vast majority of methods have studied this scenario for classification networks, where for each new task the classification layer of the network must be augmented with additional weights to make room for the newly added classes. Embedding networks have the advantage that new classes can be naturally included into the network without adding new weights. Therefore, we study incremental learning for embedding networks. In addition, we propose a new method to estimate the drift, called semantic drift, of features and compensate for it without the need of any exemplars. We approximate the drift of previous tasks based on the drift that is experienced by current task data. We perform experiments on fine-grained datasets, CIFAR100 and ImageNet-Subset. We demonstrate that embedding networks suffer significantly less from catastrophic forgetting. We outperform existing methods which do not require exemplars and obtain competitive results compared to methods which store exemplars. Furthermore, we show that our proposed SDC when combined with existing methods to prevent forgetting consistently improves results.

Image-to-image translation has recently achieved remarkable results. But despite current success, it suffers from inferior performance when translations between classes require large shape changes. We attribute this to the high-resolution bottlenecks which are used by current state-of-the-art image-to-image methods. Therefore, in this work, we propose a novel deep hierarchical Image-to-Image Translation method, called DeepI2I. We learn a model by leveraging hierarchical features: (a) structural information contained in the bottom layers and (b) semantic information extracted from the top layers. To enable the training of deep I2I models on small datasets, we propose a novel transfer learning method, that transfers knowledge from pre-trained GANs. Specifically, we leverage the discriminator of a pre-trained GANs (i.e. BigGAN or StyleGAN) to initialize both the encoder and the discriminator and the pre-trained generator to initialize the generator of our model. Applying knowledge transfer leads to an alignment problem between the encoder and generator. We introduce an adaptor network to address this. On many-class image-to-image translation on three datasets (Animal faces, Birds, and Foods) we decrease mFID by at least 35% when compared to the state-of-the-art. Furthermore, we qualitatively and quantitatively demonstrate that transfer learning significantly improves the performance of I2I systems, especially for small datasets. Finally, we are the first to perform I2I translations for domains with over 100 classes.

Metric learning networks are used to compute image embeddings, which are widely used in many applications such as image retrieval and face recognition. In this paper, we propose to use network distillation to efficiently compute image embeddings with small networks. Network distillation has been successfully applied to improve image classification, but has hardly been explored for metric learning. To do so, we propose two new loss functions that model the communication of a deep teacher network to a small student network. We evaluate our system in several datasets, including CUB-200-2011, Cars-196, Stanford Online Products and show that embeddings computed using small student networks perform significantly better than those computed using standard networks of similar size. Results on a very compact network (MobileNet-0.25), which can be used on mobile devices, show that the proposed method can greatly improve Recall@1 results from 27.5\% to 44.6\%. Furthermore, we investigate various aspects of distillation for embeddings, including hint and attention layers, semi-supervised learning and cross quality distillation. (Code is available at https://github.com/yulu0724/EmbeddingDistillation).

The majority of existing color naming methods focuses on the eleven basic color terms of the English language. However, in many applications, different sets of color names are used for the accurate description of objects. Labeling data to learn these domain-specific color names is an expensive and laborious task. Therefore, in this article we aim to learn color names from weakly labeled data. For this purpose, we add an attention branch to the color naming network. The attention branch is used to modulate the pixel-wise color naming predictions of the network. In experiments, we illustrate that the attention branch correctly identifies the relevant regions. Furthermore, we show that our method obtains state-of-the-art results for pixel-wise and image-wise classification on the EBAY dataset and is able to learn color names for various domains.