TAIL: Text-Audio Incremental Learning

Many studies combine text and audio to capture multi-modal information but they overlook the model's generalization ability on new datasets. Introducing new datasets may affect the feature space of the original dataset, leading to catastrophic forgetting. Meanwhile, large model parameters can significantly impact training performance. To address these limitations, we introduce a novel task called Text-Audio Incremental Learning (TAIL) task for text-audio retrieval, and propose a new method, PTAT, Prompt Tuning for Audio-Text incremental learning. This method utilizes prompt tuning to optimize the model parameters while incorporating an audio-text similarity and feature distillation module to effectively mitigate catastrophic forgetting. We benchmark our method and previous incremental learning methods on AudioCaps, Clotho, BBC Sound Effects and Audioset datasets, and our method outperforms previous methods significantly, particularly demonstrating stronger resistance to forgetting on older datasets. Compared to the full-parameters Finetune (Sequential) method, our model only requires 2.42\% of its parameters, achieving 4.46\% higher performance.

Facilitate Collaboration between Large Language Model and Task-specific Model for Time Series Anomaly Detection

In anomaly detection, methods based on large language models (LLMs) can incorporate expert knowledge, while task-specific smaller models excel at extracting normal patterns and detecting value fluctuations. Inspired by the human nervous system, where the brain stores expert knowledge and the peripheral nervous system and spinal cord handle specific tasks like withdrawal and knee-jerk reflexes, we propose CoLLaTe, a framework designed to facilitate collaboration between LLMs and task-specific models, leveraging the strengths of both. In this work, we first formulate the collaboration process and identify two key challenges in the collaboration between LLMs and task-specific models: (1) the misalignment between the expression domains of LLMs and smaller models, and (2) error accumulation arising from the predictions of both models. To address these challenges, we introduce two key components in CoLLaTe: the alignment module and the collaborative loss function. Through theoretical analysis and experimental validation, we demonstrate that these components effectively mitigate the identified challenges and achieve better performance than LLM based methods and task-specific smaller model.

Improving Multimodal LLMs Ability In Geometry Problem Solving, Reasoning, And Multistep Scoring

This paper presents GPSM4K, a comprehensive geometry multimodal dataset tailored to augment the problem-solving capabilities of Large Vision Language Models (LVLMs). GPSM4K encompasses 2157 multimodal question-answer pairs manually extracted from mathematics textbooks spanning grades 7-12 and is further augmented to 5340 problems, consisting of both numerical and theorem-proving questions. In contrast to PGPS9k, Geometry3K, and Geo170K which feature only objective-type questions, GPSM4K offers detailed step-by-step solutions in a consistent format, facilitating a comprehensive evaluation of problem-solving approaches. This dataset serves as an excellent benchmark for assessing the geometric reasoning capabilities of LVLMs. Evaluation of our test set shows that there is scope for improvement needed in open-source language models in geometry problem-solving. Finetuning on our training set increases the geometry problem-solving capabilities of models. Further, We also evaluate the effectiveness of techniques such as image captioning and Retrieval Augmentation generation (RAG) on model performance. We leveraged LLM to automate the task of final answer evaluation by providing ground truth and predicted solutions. This research will help to assess and improve the geometric reasoning capabilities of LVLMs.

Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts

Time series foundation models have demonstrated impressive performance as zero-shot forecasters. However, achieving effectively unified training on time series remains an open challenge. Existing approaches introduce some level of model specialization to account for the highly heterogeneous nature of time series data. For instance, Moirai pursues unified training by employing multiple input/output projection layers, each tailored to handle time series at a specific frequency. Similarly, TimesFM maintains a frequency embedding dictionary for this purpose. We identify two major drawbacks to this human-imposed frequency-level model specialization: (1) Frequency is not a reliable indicator of the underlying patterns in time series. For example, time series with different frequencies can display similar patterns, while those with the same frequency may exhibit varied patterns. (2) Non-stationarity is an inherent property of real-world time series, leading to varied distributions even within a short context window of a single time series. Frequency-level specialization is too coarse-grained to capture this level of diversity. To address these limitations, this paper introduces Moirai-MoE, using a single input/output projection layer while delegating the modeling of diverse time series patterns to the sparse mixture of experts (MoE) within Transformers. With these designs, Moirai-MoE reduces reliance on human-defined heuristics and enables automatic token-level specialization. Extensive experiments on 39 datasets demonstrate the superiority of Moirai-MoE over existing foundation models in both in-distribution and zero-shot scenarios. Furthermore, this study conducts comprehensive model analyses to explore the inner workings of time series MoE foundation models and provides valuable insights for future research.

Manifold-Aware Local Feature Modeling for Semi-Supervised Medical Image Segmentation

Achieving precise medical image segmentation is vital for effective treatment planning and accurate disease diagnosis. Traditional fully-supervised deep learning methods, though highly precise, are heavily reliant on large volumes of labeled data, which are often difficult to obtain due to the expertise required for medical annotations. This has led to the rise of semi-supervised learning approaches that utilize both labeled and unlabeled data to mitigate the label scarcity issue. In this paper, we introduce the Manifold-Aware Local Feature Modeling Network (MANet), which enhances the U-Net architecture by incorporating manifold supervision signals. This approach focuses on improving boundary accuracy, which is crucial for reliable medical diagnosis. To further extend the versatility of our method, we propose two variants: MA-Sobel and MA-Canny. The MA-Sobel variant employs the Sobel operator, which is effective for both 2D and 3D data, while the MA-Canny variant utilizes the Canny operator, specifically designed for 2D images, to refine boundary detection. These variants allow our method to adapt to various medical image modalities and dimensionalities, ensuring broader applicability. Our extensive experiments on datasets such as ACDC, LA, and Pancreas-NIH demonstrate that MANet consistently surpasses state-of-the-art methods in performance metrics like Dice and Jaccard scores. The proposed method also shows improved generalization across various semi-supervised segmentation networks, highlighting its robustness and effectiveness. Visual analysis of segmentation results confirms that MANet offers clearer and more accurate class boundaries, underscoring the value of manifold information in medical image segmentation.

Grounding is All You Need? Dual Temporal Grounding for Video Dialog

In the realm of video dialog response generation, the understanding of video content and the temporal nuances of conversation history are paramount. While a segment of current research leans heavily on large-scale pretrained visual-language models and often overlooks temporal dynamics, another delves deep into spatial-temporal relationships within videos but demands intricate object trajectory pre-extractions and sidelines dialog temporal dynamics. This paper introduces the Dual Temporal Grounding-enhanced Video Dialog model (DTGVD), strategically designed to merge the strengths of both dominant approaches. It emphasizes dual temporal relationships by predicting dialog turn-specific temporal regions, filtering video content accordingly, and grounding responses in both video and dialog contexts. One standout feature of DTGVD is its heightened attention to chronological interplay. By recognizing and acting upon the dependencies between different dialog turns, it captures more nuanced conversational dynamics. To further bolster the alignment between video and dialog temporal dynamics, we've implemented a list-wise contrastive learning strategy. Within this framework, accurately grounded turn-clip pairings are designated as positive samples, while less precise pairings are categorized as negative. This refined classification is then funneled into our holistic end-to-end response generation mechanism. Evaluations using AVSD@DSTC-7 and AVSD@DSTC-8 datasets underscore the superiority of our methodology.

DriveDiTFit: Fine-tuning Diffusion Transformers for Autonomous Driving

In autonomous driving, deep models have shown remarkable performance across various visual perception tasks with the demand of high-quality and huge-diversity training datasets. Such datasets are expected to cover various driving scenarios with adverse weather, lighting conditions and diverse moving objects. However, manually collecting these data presents huge challenges and expensive cost. With the rapid development of large generative models, we propose DriveDiTFit, a novel method for efficiently generating autonomous Driving data by Fine-tuning pre-trained Diffusion Transformers (DiTs). Specifically, DriveDiTFit utilizes a gap-driven modulation technique to carefully select and efficiently fine-tune a few parameters in DiTs according to the discrepancy between the pre-trained source data and the target driving data. Additionally, DriveDiTFit develops an effective weather and lighting condition embedding module to ensure diversity in the generated data, which is initialized by a nearest-semantic-similarity initialization approach. Through progressive tuning scheme to refined the process of detail generation in early diffusion process and enlarging the weights corresponding to small objects in training loss, DriveDiTFit ensures high-quality generation of small moving objects in the generated data. Extensive experiments conducted on driving datasets confirm that our method could efficiently produce diverse real driving data. The source codes will be available at https://github.com/TtuHamg/DriveDiTFit.

Described Spatial-Temporal Video Detection

Detecting visual content on language expression has become an emerging topic in the community. However, in the video domain, the existing setting, i.e., spatial-temporal video grounding (STVG), is formulated to only detect one pre-existing object in each frame, ignoring the fact that language descriptions can involve none or multiple entities within a video. In this work, we advance the STVG to a more practical setting called described spatial-temporal video detection (DSTVD) by overcoming the above limitation. To facilitate the exploration of DSTVD, we first introduce a new benchmark, namely DVD-ST. Notably, DVD-ST supports grounding from none to many objects onto the video in response to queries and encompasses a diverse range of over 150 entities, including appearance, actions, locations, and interactions. The extensive breadth and diversity of the DVD-ST dataset make it an exemplary testbed for the investigation of DSTVD. In addition to the new benchmark, we further present two baseline methods for our proposed DSTVD task by extending two representative STVG models, i.e., TubeDETR, and STCAT. These extended models capitalize on tubelet queries to localize and track referred objects across the video sequence. Besides, we adjust the training objectives of these models to optimize spatial and temporal localization accuracy and multi-class classification capabilities. Furthermore, we benchmark the baselines on the introduced DVD-ST dataset and conduct extensive experimental analysis to guide future investigation. Our code and benchmark will be publicly available.

Do As I Do: Pose Guided Human Motion Copy

Human motion copy is an intriguing yet challenging task in artificial intelligence and computer vision, which strives to generate a fake video of a target person performing the motion of a source person. The problem is inherently challenging due to the subtle human-body texture details to be generated and the temporal consistency to be considered. Existing approaches typically adopt a conventional GAN with an L1 or L2 loss to produce the target fake video, which intrinsically necessitates a large number of training samples that are challenging to acquire. Meanwhile, current methods still have difficulties in attaining realistic image details and temporal consistency, which unfortunately can be easily perceived by human observers. Motivated by this, we try to tackle the issues from three aspects: (1) We constrain pose-to-appearance generation with a perceptual loss and a theoretically motivated Gromov-Wasserstein loss to bridge the gap between pose and appearance. (2) We present an episodic memory module in the pose-to-appearance generation to propel continuous learning that helps the model learn from its past poor generations. We also utilize geometrical cues of the face to optimize facial details and refine each key body part with a dedicated local GAN. (3) We advocate generating the foreground in a sequence-to-sequence manner rather than a single-frame manner, explicitly enforcing temporal inconsistency. Empirical results on five datasets, iPER, ComplexMotion, SoloDance, Fish, and Mouse datasets, demonstrate that our method is capable of generating realistic target videos while precisely copying motion from a source video. Our method significantly outperforms state-of-the-art approaches and gains 7.2% and 12.4% improvements in PSNR and FID respectively.

PetalView: Fine-grained Location and Orientation Extraction of Street-view Images via Cross-view Local Search with Supplementary Materials

Satellite-based street-view information extraction by cross-view matching refers to a task that extracts the location and orientation information of a given street-view image query by using one or multiple geo-referenced satellite images. Recent work has initiated a new research direction to find accurate information within a local area covered by one satellite image centered at a location prior (e.g., from GPS). It can be used as a standalone solution or complementary step following a large-scale search with multiple satellite candidates. However, these existing works require an accurate initial orientation (angle) prior (e.g., from IMU) and/or do not efficiently search through all possible poses. To allow efficient search and to give accurate prediction regardless of the existence or the accuracy of the angle prior, we present PetalView extractors with multi-scale search. The PetalView extractors give semantically meaningful features that are equivalent across two drastically different views, and the multi-scale search strategy efficiently inspects the satellite image from coarse to fine granularity to provide sub-meter and sub-degree precision extraction. Moreover, when an angle prior is given, we propose a learnable prior angle mixer to utilize this information. Our method obtains the best performance on the VIGOR dataset and successfully improves the performance on KITTI dataset test 1 set with the recall within 1 meter (r@1m) for location estimation to 68.88% and recall within 1 degree (r@1d) 21.10% when no angle prior is available, and with angle prior achieves stable estimations at r@1m and r@1d above 70% and 21%, up to a 40-degree noise level.