Image-goal navigation aims to steer an agent towards the goal location specified by an image. Most prior methods tackle this task by learning a navigation policy, which extracts visual features of goal and observation images, compares their similarity and predicts actions. However, if the agent is in a different room from the goal image, it's extremely challenging to identify their similarity and infer the likely goal location, which may result in the agent wandering around. Intuitively, when humans carry out this task, they may roughly compare the current observation with the goal image, having an approximate concept of whether they are in the same room before executing the actions. Inspired by this intuition, we try to imitate human behaviour and propose a Room Expert Guided Image-Goal Navigation model~(REGNav) to equip the agent with the ability to analyze whether goal and observation images are taken in the same room. Specifically, we first pre-train a room expert with an unsupervised learning technique on the self-collected unlabelled room images. The expert can extract the hidden room style information of goal and observation images and predict their relationship about whether they belong to the same room. In addition, two different fusion approaches are explored to efficiently guide the agent navigation with the room relation knowledge. Extensive experiments show that our REGNav surpasses prior state-of-the-art works on three popular benchmarks.

Dense video captioning aims to localize and caption all events in arbitrary untrimmed videos. Although previous methods have achieved appealing results, they still face the issue of temporal bias, i.e, models tend to focus more on events with certain temporal characteristics. Specifically, 1) the temporal distribution of events in training datasets is uneven. Models trained on these datasets will pay less attention to out-of-distribution events. 2) long-duration events have more frame features than short ones and will attract more attention. To address this, we argue that events, with varying temporal characteristics, should be treated equally when it comes to dense video captioning. Intuitively, different events tend to have distinct visual differences due to varied camera views, backgrounds, or subjects. Inspired by that, we intend to utilize visual features to have an approximate perception of possible events and pay equal attention to them. In this paper, we introduce a simple but effective framework, called Event-Equalized Dense Video Captioning (E2DVC) to overcome the temporal bias and treat all possible events equally. Experimental results on ActivityNet Captions and YouCook2 dataset validate the effectiveness of the proposed methods and show State-of-the-art (SOTA) performance on dense video captioning.

Source-free domain generalization (SFDG) tackles the challenge of adapting models to unseen target domains without access to source domain data. To deal with this challenging task, recent advances in SFDG have primarily focused on leveraging the text modality of vision-language models such as CLIP. These methods involve developing a transferable linear classifier based on diverse style features extracted from the text and learned prompts or deriving domain-unified text representations from domain banks. However, both style features and domain banks have limitations in capturing comprehensive domain knowledge. In this work, we propose Prompt-Driven Text Adapter (PromptTA) method, which is designed to better capture the distribution of style features and employ resampling to ensure thorough coverage of domain knowledge. To further leverage this rich domain information, we introduce a text adapter that learns from these style features for efficient domain information storage. Extensive experiments conducted on four benchmark datasets demonstrate that PromptTA achieves state-of-the-art performance. The code is available at https://github.com/zhanghr2001/PromptTA.

Leveraging the vision foundation models has emerged as a mainstream paradigm that improves the performance of image feature matching. However, previous works have ignored the misalignment when introducing the foundation models into feature matching. The misalignment arises from the discrepancy between the foundation models focusing on single-image understanding and the cross-image understanding requirement of feature matching. Specifically, 1) the embeddings derived from commonly used foundation models exhibit discrepancies with the optimal embeddings required for feature matching; 2) lacking an effective mechanism to leverage the single-image understanding ability into cross-image understanding. A significant consequence of the misalignment is they struggle when addressing multi-instance feature matching problems. To address this, we introduce a simple but effective framework, called IMD (Image feature Matching with a pre-trained Diffusion model) with two parts: 1) Unlike the dominant solutions employing contrastive-learning based foundation models that emphasize global semantics, we integrate the generative-based diffusion models to effectively capture instance-level details. 2) We leverage the prompt mechanism in generative model as a natural tunnel, propose a novel cross-image interaction prompting module to facilitate bidirectional information interaction between image pairs. To more accurately measure the misalignment, we propose a new benchmark called IMIM, which focuses on multi-instance scenarios. Our proposed IMD establishes a new state-of-the-art in commonly evaluated benchmarks, and the superior improvement 12% in IMIM indicates our method efficiently mitigates the misalignment.

Distributional Compositional Generalization (DCG) refers to the ability to tackle tasks from new distributions by leveraging the knowledge of concepts learned from supporting distributions. In this work, we aim to explore the statistical mechanisms of DCG, which have been largely overlooked in previous studies. By statistically formulating the problem, this paper seeks to address two key research questions: 1) Can a method to one DCG problem be applicable to another? 2) What statistical properties can indicate a learning algorithm's capacity for knowledge composition in DCG tasks? \textbf{To address the first question}, an invariant measure is proposed to provide a dimension where all different methods converge. This measure underscores the critical role of data in enabling improvements without trade-offs. \textbf{As for the second question}, we reveal that by decoupling the impacts of insufficient data and knowledge composition, the ability of the learning algorithm to compose knowledge relies on the compatibility and sensitivity between the learning algorithm and the composition rule. In summary, the statistical analysis of the generalization mechanisms provided in this paper deepens our understanding of compositional generalization, offering a complementary evidence on the importance of data in DCG task.

Visual place recognition is a challenging task for autonomous driving and robotics, which is usually considered as an image retrieval problem. A commonly used two-stage strategy involves global retrieval followed by re-ranking using patch-level descriptors. Most deep learning-based methods in an end-to-end manner cannot extract global features with sufficient semantic information from RGB images. In contrast, re-ranking can utilize more explicit structural and semantic information in one-to-one matching process, but it is time-consuming. To bridge the gap between global retrieval and re-ranking and achieve a good trade-off between accuracy and efficiency, we propose StructVPR++, a framework that embeds structural and semantic knowledge into RGB global representations via segmentation-guided distillation. Our key innovation lies in decoupling label-specific features from global descriptors, enabling explicit semantic alignment between image pairs without requiring segmentation during deployment. Furthermore, we introduce a sample-wise weighted distillation strategy that prioritizes reliable training pairs while suppressing noisy ones. Experiments on four benchmarks demonstrate that StructVPR++ surpasses state-of-the-art global methods by 5-23% in Recall@1 and even outperforms many two-stage approaches, achieving real-time efficiency with a single RGB input.

The rapid development of deep-learning enabled task-oriented communications (TOC) significantly shifts the paradigm of wireless communications. However, the high computation demands, particularly in resource-constrained systems e.g., mobile phones and UAVs, make TOC challenging for many tasks. To address the problem, we propose a novel TOC method with two models: a static and a dynamic model. In the static model, we apply a neural network (NN) as a task-oriented encoder (TOE) when there is no computation budget constraint. The dynamic model is used when device computation resources are limited, and it uses dynamic NNs with multiple exits as the TOE. The dynamic model sorts input data by complexity with thresholds, allowing the efficient allocation of computation resources. Furthermore, we analyze the convergence of the proposed TOC methods and show that the model converges at rate$O\left ({{\frac {1}{\sqrt {T}}}}\right)$with an epoch of lengthT. Experimental results demonstrate that the static model outperforms baseline models in terms of transmitted dimensions, floating-point operations (FLOPs), and accuracy simultaneously. The dynamic model can further improve accuracy and computational demand, providing an improved solution for resource-constrained systems.

Aiming to estimate the location, a common strategy of Visual Place Recognition (VPR) involves utilizing global retrieval to get top-k candidates first and performing local feature matching in candidates for reranking. Although local reranking methods bring performance gains, they need a lot of computational overhead. To narrow the performance gap between global retrieval and local reranking methods with little cost, one method is to rerank candidates with global features. However, previous works only utilized the information from positive samples in candidates, ignoring the fact that negative samples can also provide useful information. To this end, we propose RankTuning, a method that aggregates all the information from candidates using global features for reranking. Specifically, we design a cross-image interaction module that allows all candidates to interact with others to enhance the discriminative power of features. Furthermore, to drive the training of this module, we propose Generalized Recall loss to handle hard samples with a better gradient strategy. Experimental results demonstrate that our method can be easily inserted into existing architectures and achieve state-of-the-art performance. Meanwhile, our method does not require additional storage overhead, and the matching latency is only 6.3% of that of the current fastest local reranking method. The code is released athttps://github.com/LKELN/RankTuning.git

In-context learning, i.e., learning from context examples, is an impressive ability of Transformer. Training Transformers to possess this in-context learning skill is computationally intensive due to the occurrence of *learning plateaus*, which are periods within the training process where there is minimal or no enhancement in the model's in-context learning capability. To study the mechanism behind the learning plateaus, we conceptually separate a component within the model's internal representation that is exclusively affected by the model's weights. We call this the “weights component”, and the remainder is identified as the “context component”. By conducting meticulous and controlled experiments on synthetic tasks, we note that the persistence of learning plateaus correlates with compromised functionality of the weights component. Recognizing the impaired performance of the weights component as a fundamental behavior that drives learning plateaus, we have developed three strategies to expedite the learning of Transformers. The effectiveness of these strategies is further confirmed in natural language processing tasks. In conclusion, our research demonstrates the feasibility of cultivating a powerful in-context learning ability within AI systems in an eco-friendly manner.

Trajectory prediction plays a crucial role in autonomous driving, but it is challenging due to the multi-modal nature of future trajectories. Behavior information is frequently employed to capture more diverse modalities of future trajectories. Traditional behavior information is typically hard-encoded, which is often inaccurate and inadequate for reflecting future multimodality. Therefore, we introduce the concept of soft vehicle behavior, which is represented as a probability distribution over a predefined comprehensive set of behaviors. This approach allows for a more rational depiction of vehicle behavior and captures potential future driving modalities. Based on it, we propose a new soft-behavior-constrained vehicle trajectory prediction framework. The framework consists of a backbone and a lightweight and plug-and-play behavior prediction module, which is used to imbue soft behavior constraints to assist in representation learning. We integrated the behavior prediction module into five representative trajectory predictors and achieved improvements of at least 4.2% in minFDE(K=5) on the nuScenes dataset and 0.5% in minFDE(K=6) on the Argoverse 1 motion forecasting dataset. These universal increments prove the effectiveness and generalizability of soft behavior constraints in vehicle trajectory prediction.

Homography estimation is the task of determining the transformation from an image pair. Our approach focuses on employing detector-free feature matching methods to address this issue. Previous work has underscored the importance of incorporating semantic information, however there still lacks an efficient way to utilize semantic information. Previous methods suffer from treating the semantics as a pre-processing, causing the utilization of semantics overly coarse-grained and lack adaptability when dealing with different tasks. In our work, we seek another way to use the semantic information, that is semantic-aware feature representation learning framework. Based on this, we propose SRMatcher, a new detector-free feature matching method, which encourages the network to learn integrated semantic feature representation. Specifically, to capture precise and rich semantics, we leverage the capabilities of recently popularized vision foundation models (VFMs) trained on extensive datasets. Then, a cross-images Semantic-aware Fusion Block (SFB) is proposed to integrate its fine-grained semantic features into the feature representation space. In this way, by reducing errors stemming from semantic inconsistencies in matching pairs, our proposed SRMatcher is able to deliver more accurate and realistic outcomes. Extensive experiments show that SRMatcher surpasses solid baselines and attains SOTA results on multiple real-world datasets. Compared to the previous SOTA approach GeoFormer, SRMatcher increases the area under the cumulative curve (AUC) by about 11% on HPatches. Additionally, the SRMatcher could serve as a plug-and-play framework for other matching methods like LoFTR, yielding substantial precision improvement.

Visual place recognition (VPR) is usually considered as a specific image retrieval problem. Limited by existing training frameworks, most deep learning-based works cannot extract sufficiently stable global features from RGB images and rely on a time-consuming re-ranking step to exploit spatial structural information for better performance. In this paper, we propose StructVPR, a novel training architecture for VPR, to enhance structural knowledge in RGB global features and thus improve feature stability in a constantly changing environment. Specifically, StructVPR uses segmentation images as a more definitive source of structural knowledge input into a CNN network and applies knowledge distillation to avoid online segmentation and inference of seg-branch in testing. Considering that not all samples contain high-quality and helpful knowledge, and some even hurt the performance of distillation, we partition samples and weigh each sample's distillation loss to enhance the expected knowledge precisely. Finally, StructVPR achieves impressive performance on several benchmarks using only global retrieval and even outperforms many two-stage approaches by a large margin. After adding additional re-ranking, ours achieves state-of-the-art performance while maintaining a low computational cost.

This paper explores the connection between learning trajectories of Deep Neural Networks (DNNs) and their generalization capabilities when optimized using (stochastic) gradient descent algorithms. Instead of concentrating solely on the generalization error of the DNN post-training, we present a novel perspective for analyzing generalization error by investigating the contribution of each update step to the change in generalization error. This perspective enable a more direct comprehension of how the learning trajectory influences generalization error. Building upon this analysis, we propose a new generalization bound that incorporates more extensive trajectory information. Our proposed generalization bound depends on the complexity of learning trajectory and the ratio between the bias and diversity of training set. Experimental observations reveal that our method effectively captures the generalization error throughout the training process. Furthermore, our approach can also track changes in generalization error when adjustments are made to learning rates and label noise levels. These results demonstrate that learning trajectory information is a valuable indicator of a model's generalization capabilities.

Recently, Arjevani et al. [1] establish a lower bound of iteration complexity for the first-order optimization under an $L$-smooth condition and a bounded noise variance assumption. However, a thorough review of existing literature on Adam's convergence reveals a noticeable gap: none of them meet the above lower bound. In this paper, we close the gap by deriving a new convergence guarantee of Adam, with only an $L$-smooth condition and a bounded noise variance assumption. Our results remain valid across a broad spectrum of hyperparameters. Especially with properly chosen hyperparameters, we derive an upper bound of the iteration complexity of Adam and show that it meets the lower bound for first-order optimizers. To the best of our knowledge, this is the first to establish such a tight upper bound for Adam's convergence. Our proof utilizes novel techniques to handle the entanglement between momentum and adaptive learning rate and to convert the first-order term in the Descent Lemma to the gradient norm, which may be of independent interest.

With a large number of distance measures, the appropriate choice for clustering a given data set with a specified clustering algorithm becomes an important problem. In this article, an automatic distance measure recommendation method for clustering algorithms is proposed. The recommendation method consists of the following steps: (1) metadata extraction, including meta-feature collection and meta-target identification; (2) recommendation model construction using metadata; and (3) distance measure recommendation for a new data set by the recommendation model. Two different types of meta-targets and meta-learning techniques are utilized considering the possible different requirements of users. To validate the necessity and effectiveness of the distance measure recommendation method, an empirical study is conducted with 199 publicly available data sets, 9 distance measures, and 2 widely used clustering algorithms. The experimental results indicate that distance measure significantly influences the performance of the clustering algorithm for a given data set. Furthermore, performance analysis of the proposed recommendation method proves its effectiveness.