Abstract:Enabling Large Language Models (LLMs) to understand the 3D physical world is an emerging yet challenging research direction. Current strategies for processing point clouds typically downsample the scene or divide it into smaller parts for separate analysis. However, both approaches risk losing key local details or global contextual information. In this paper, we introduce PerLA, a 3D language assistant designed to be more perceptive to both details and context, making visual representations more informative for the LLM. PerLA captures high-resolution (local) details in parallel from different point cloud areas and integrates them with (global) context obtained from a lower-resolution whole point cloud. We present a novel algorithm that preserves point cloud locality through the Hilbert curve and effectively aggregates local-to-global information via cross-attention and a graph neural network. Lastly, we introduce a novel loss for local representation consensus to promote training stability. PerLA outperforms state-of-the-art 3D language assistants, with gains of up to +1.34 CiDEr on ScanQA for question answering, and +4.22 on ScanRefer and +3.88 on Nr3D for dense captioning.\url{https://gfmei.github.io/PerLA/}
Abstract:Tuning effective step sizes is crucial for the stability and efficiency of optimization algorithms. While adaptive coordinate-wise step sizes tuning methods have been explored in first-order methods, second-order methods still lack efficient techniques. Current approaches, including hypergradient descent and cutting plane methods, offer limited improvements or encounter difficulties in second-order contexts. To address these challenges, we introduce a novel Learning-to-Optimize (L2O) model within the Broyden-Fletcher-Goldfarb-Shanno (BFGS) framework, which leverages neural networks to predict optimal coordinate-wise step sizes. Our model integrates a theoretical foundation that establishes conditions for the stability and convergence of these step sizes. Extensive experiments demonstrate that our approach achieves substantial improvements over traditional backtracking line search and hypergradient descent-based methods, offering up to 7$\times$ faster and stable performance across diverse optimization tasks.
Abstract:Vision-Language Models (VLMs) have shown remarkable capabilities across diverse visual tasks, including image recognition, video understanding, and Visual Question Answering (VQA) when explicitly trained for these tasks. Despite these advances, we find that current VLMs lack a fundamental cognitive ability: learning to localize objects in a scene by taking into account the context. In this work, we focus on the task of few-shot personalized localization, where a model is given a small set of annotated images (in-context examples) -- each with a category label and bounding box -- and is tasked with localizing the same object type in a query image. To provoke personalized localization abilities in models, we present a data-centric solution that fine-tunes them using carefully curated data from video object tracking datasets. By leveraging sequences of frames tracking the same object across multiple shots, we simulate instruction-tuning dialogues that promote context awareness. To reinforce this, we introduce a novel regularization technique that replaces object labels with pseudo-names, ensuring the model relies on visual context rather than prior knowledge. Our method significantly enhances few-shot localization performance without sacrificing generalization, as demonstrated on several benchmarks tailored to personalized localization. This work is the first to explore and benchmark personalized few-shot localization for VLMs, laying a foundation for future research in context-driven vision-language applications. The code for our project is available at https://github.com/SivanDoveh/IPLoc
Abstract:Complex systems in physics, chemistry, and biology that evolve over time with inherent randomness are typically described by stochastic differential equations (SDEs). A fundamental challenge in science and engineering is to determine the governing equations of a complex system from snapshot data. Traditional equation discovery methods often rely on stringent assumptions, such as the availability of the trajectory information or time-series data, and the presumption that the underlying system is deterministic. In this work, we introduce a data-driven, simulation-free framework, called Sparse Identification of Differential Equations from Snapshots (SpIDES), that discovers the governing equations of a complex system from snapshots by utilizing the advanced machine learning techniques to perform three essential steps: probability flow reconstruction, probability density estimation, and Bayesian sparse identification. We validate the effectiveness and robustness of SpIDES by successfully identifying the governing equation of an over-damped Langevin system confined within two potential wells. By extracting interpretable drift and diffusion terms from the SDEs, our framework provides deeper insights into system dynamics, enhances predictive accuracy, and facilitates more effective strategies for managing and simulating stochastic systems.
Abstract:Convolutional neural networks (CNNs) and vision transformers (ViTs) have become essential in computer vision for local and global feature extraction. However, aggregating these architectures in existing methods often results in inefficiencies. To address this, the CNN-Transformer Aggregation Network (CTA-Net) was developed. CTA-Net combines CNNs and ViTs, with transformers capturing long-range dependencies and CNNs extracting localized features. This integration enables efficient processing of detailed local and broader contextual information. CTA-Net introduces the Light Weight Multi-Scale Feature Fusion Multi-Head Self-Attention (LMF-MHSA) module for effective multi-scale feature integration with reduced parameters. Additionally, the Reverse Reconstruction CNN-Variants (RRCV) module enhances the embedding of CNNs within the transformer architecture. Extensive experiments on small-scale datasets with fewer than 100,000 samples show that CTA-Net achieves superior performance (TOP-1 Acc 86.76\%), fewer parameters (20.32M), and greater efficiency (FLOPs 2.83B), making it a highly efficient and lightweight solution for visual tasks on small-scale datasets (fewer than 100,000).
Abstract:The large-scale training of multi-modal models on data scraped from the web has shown outstanding utility in infusing these models with the required world knowledge to perform effectively on multiple downstream tasks. However, one downside of scraping data from the web can be the potential sacrifice of the benchmarks on which the abilities of these models are often evaluated. To safeguard against test data contamination and to truly test the abilities of these foundation models we propose LiveXiv: A scalable evolving live benchmark based on scientific ArXiv papers. LiveXiv accesses domain-specific manuscripts at any given timestamp and proposes to automatically generate visual question-answer pairs (VQA). This is done without any human-in-the-loop, using the multi-modal content in the manuscripts, like graphs, charts, and tables. Moreover, we introduce an efficient evaluation approach that estimates the performance of all models on the evolving benchmark using evaluations of only a subset of models. This significantly reduces the overall evaluation cost. We benchmark multiple open and proprietary Large Multi-modal Models (LMMs) on the first version of our benchmark, showing its challenging nature and exposing the models true abilities, avoiding contamination. Lastly, in our commitment to high quality, we have collected and evaluated a manually verified subset. By comparing its overall results to our automatic annotations, we have found that the performance variance is indeed minimal (<2.5%). Our dataset is available online on HuggingFace, and our code will be available here.
Abstract:In this work, we propose a novel method (GLOV) enabling Large Language Models (LLMs) to act as implicit Optimizers for Vision-Langugage Models (VLMs) to enhance downstream vision tasks. Our GLOV meta-prompts an LLM with the downstream task description, querying it for suitable VLM prompts (e.g., for zero-shot classification with CLIP). These prompts are ranked according to a purity measure obtained through a fitness function. In each respective optimization step, the ranked prompts are fed as in-context examples (with their accuracies) to equip the LLM with the knowledge of the type of text prompts preferred by the downstream VLM. Furthermore, we also explicitly steer the LLM generation process in each optimization step by specifically adding an offset difference vector of the embeddings from the positive and negative solutions found by the LLM, in previous optimization steps, to the intermediate layer of the network for the next generation step. This offset vector steers the LLM generation toward the type of language preferred by the downstream VLM, resulting in enhanced performance on the downstream vision tasks. We comprehensively evaluate our GLOV on 16 diverse datasets using two families of VLMs, i.e., dual-encoder (e.g., CLIP) and encoder-decoder (e.g., LLaVa) models -- showing that the discovered solutions can enhance the recognition performance by up to 15.0% and 57.5% (3.8% and 21.6% on average) for these models.
Abstract:Click-Through Rate (CTR) prediction is crucial for Recommendation System(RS), aiming to provide personalized recommendation services for users in many aspects such as food delivery, e-commerce and so on. However, traditional RS relies on collaborative signals, which lacks semantic understanding to real-time scenes. We also noticed that a major challenge in utilizing Large Language Models (LLMs) for practical recommendation purposes is their efficiency in dealing with long text input. To break through the problems above, we propose Large Language Model Aided Real-time Scene Recommendation(LARR), adopt LLMs for semantic understanding, utilizing real-time scene information in RS without requiring LLM to process the entire real-time scene text directly, thereby enhancing the efficiency of LLM-based CTR modeling. Specifically, recommendation domain-specific knowledge is injected into LLM and then RS employs an aggregation encoder to build real-time scene information from separate LLM's outputs. Firstly, a LLM is continual pretrained on corpus built from recommendation data with the aid of special tokens. Subsequently, the LLM is fine-tuned via contrastive learning on three kinds of sample construction strategies. Through this step, LLM is transformed into a text embedding model. Finally, LLM's separate outputs for different scene features are aggregated by an encoder, aligning to collaborative signals in RS, enhancing the performance of recommendation model.
Abstract:Dynamic activation (DA) techniques, such as DejaVu and MoEfication, have demonstrated their potential to significantly enhance the inference efficiency of large language models (LLMs). However, these techniques often rely on ReLU activation functions or require additional parameters and training to maintain performance. This paper introduces a training-free Threshold-based Dynamic Activation(TDA) method that leverage sequence information to exploit the inherent sparsity of models across various architectures. This method is designed to accelerate generation speed by 18-25\% without significantly compromising task performance, thereby addressing the limitations of existing DA techniques. Moreover, we delve into the root causes of LLM sparsity and theoretically analyze two of its critical features: history-related activation uncertainty and semantic-irrelevant activation inertia. Our comprehensive analyses not only provide a robust theoretical foundation for DA methods but also offer valuable insights to guide future research in optimizing LLMs for greater efficiency and effectiveness.
Abstract:Accurate 3D object detection in LiDAR point clouds is crucial for autonomous driving systems. To achieve state-of-the-art performance, the supervised training of detectors requires large amounts of human-annotated data, which is expensive to obtain and restricted to predefined object categories. To mitigate manual labeling efforts, recent unsupervised object detection approaches generate class-agnostic pseudo-labels for moving objects, subsequently serving as supervision signal to bootstrap a detector. Despite promising results, these approaches do not provide class labels or generalize well to static objects. Furthermore, they are mostly restricted to data containing multiple drives from the same scene or images from a precisely calibrated and synchronized camera setup. To overcome these limitations, we propose a vision-language-guided unsupervised 3D detection approach that operates exclusively on LiDAR point clouds. We transfer CLIP knowledge to classify point clusters of static and moving objects, which we discover by exploiting the inherent spatio-temporal information of LiDAR point clouds for clustering, tracking, as well as box and label refinement. Our approach outperforms state-of-the-art unsupervised 3D object detectors on the Waymo Open Dataset ($+23~\text{AP}_{3D}$) and Argoverse 2 ($+7.9~\text{AP}_{3D}$) and provides class labels not solely based on object size assumptions, marking a significant advancement in the field.