Recent advancements in multimodal models highlight the value of rewritten captions for improving performance, yet key challenges remain. For example, while synthetic captions often provide superior quality and image-text alignment, it is not clear whether they can fully replace AltTexts: the role of synthetic captions and their interaction with original web-crawled AltTexts in pre-training is still not well understood. Moreover, different multimodal foundation models may have unique preferences for specific caption formats, but efforts to identify the optimal captions for each model remain limited. In this work, we propose a novel, controllable, and scalable captioning pipeline designed to generate diverse caption formats tailored to various multimodal models. By examining Short Synthetic Captions (SSC) towards Dense Synthetic Captions (DSC+) as case studies, we systematically explore their effects and interactions with AltTexts across models such as CLIP, multimodal LLMs, and diffusion models. Our findings reveal that a hybrid approach that keeps both synthetic captions and AltTexts can outperform the use of synthetic captions alone, improving both alignment and performance, with each model demonstrating preferences for particular caption formats. This comprehensive analysis provides valuable insights into optimizing captioning strategies, thereby advancing the pre-training of multimodal foundation models.
相關內容
The increasing use of complex and opaque black box models requires the adoption of interpretable measures, one such option is extractive rationalizing models, which serve as a more interpretable alternative. These models, also known as Explain-Then-Predict models, employ an explainer model to extract rationales and subsequently condition the predictor with the extracted information. Their primary objective is to provide precise and faithful explanations, represented by the extracted rationales. In this paper, we take a semi-supervised approach to optimize for the plausibility of extracted rationales. We adopt a pre-trained natural language inference (NLI) model and further fine-tune it on a small set of supervised rationales ($10\%$). The NLI predictor is leveraged as a source of supervisory signals to the explainer via entailment alignment. We show that, by enforcing the alignment agreement between the explanation and answer in a question-answering task, the performance can be improved without access to ground truth labels. We evaluate our approach on the ERASER dataset and show that our approach achieves comparable results with supervised extractive models and outperforms unsupervised approaches by $> 100\%$.
Although most existing multi-modal salient object detection (SOD) methods demonstrate effectiveness through training models from scratch, the limited multi-modal data hinders these methods from reaching optimality. In this paper, we propose a novel framework to explore and exploit the powerful feature representation and zero-shot generalization ability of the pre-trained Segment Anything Model (SAM) for multi-modal SOD. Despite serving as a recent vision fundamental model, driving the class-agnostic SAM to comprehend and detect salient objects accurately is non-trivial, especially in challenging scenes. To this end, we develop \underline{SAM} with se\underline{m}antic f\underline{e}ature fu\underline{s}ion guidanc\underline{e} (Sammese), which incorporates multi-modal saliency-specific knowledge into SAM to adapt SAM to multi-modal SOD tasks. However, it is difficult for SAM trained on single-modal data to directly mine the complementary benefits of multi-modal inputs and comprehensively utilize them to achieve accurate saliency prediction. To address these issues, we first design a multi-modal complementary fusion module to extract robust multi-modal semantic features by integrating information from visible and thermal or depth image pairs. Then, we feed the extracted multi-modal semantic features into both the SAM image encoder and mask decoder for fine-tuning and prompting, respectively. Specifically, in the image encoder, a multi-modal adapter is proposed to adapt the single-modal SAM to multi-modal information. In the mask decoder, a semantic-geometric prompt generation strategy is proposed to produce corresponding embeddings with various saliency cues. Extensive experiments on both RGB-D and RGB-T SOD benchmarks show the effectiveness of the proposed framework. The code will be available at \url{//github.com/Angknpng/Sammese}.
The rise of delay-sensitive yet computing-intensive Internet of Things (IoT) applications poses challenges due to the limited processing power of IoT devices. Mobile Edge Computing (MEC) offers a promising solution to address these challenges by placing computing servers close to end users. Despite extensive research on MEC, optimizing network topology to improve computational efficiency remains underexplored. Recognizing the critical role of network topology, we introduce a novel decentralized network topology design strategy for task offloading (DNTD-TO) that jointly considers topology design and task allocation. Inspired by communication and sensor networks, DNTD-TO efficiently constructs three-layered network structures for task offloading and generates optimal task allocations for these structures. Comparisons with existing topology design methods demonstrate the promising performance of our approach.
Audio-visual pre-trained models have gained substantial attention recently and demonstrated superior performance on various audio-visual tasks. This study investigates whether pre-trained audio-visual models demonstrate non-arbitrary associations between sounds and visual representations$\unicode{x2013}$known as sound symbolism$\unicode{x2013}$which is also observed in humans. We developed a specialized dataset with synthesized images and audio samples and assessed these models using a non-parametric approach in a zero-shot setting. Our findings reveal a significant correlation between the models' outputs and established patterns of sound symbolism, particularly in models trained on speech data. These results suggest that such models can capture sound-meaning connections akin to human language processing, providing insights into both cognitive architectures and machine learning strategies.
We present Deep Compression Autoencoder (DC-AE), a new family of autoencoder models for accelerating high-resolution diffusion models. Existing autoencoder models have demonstrated impressive results at a moderate spatial compression ratio (e.g., 8x), but fail to maintain satisfactory reconstruction accuracy for high spatial compression ratios (e.g., 64x). We address this challenge by introducing two key techniques: (1) Residual Autoencoding, where we design our models to learn residuals based on the space-to-channel transformed features to alleviate the optimization difficulty of high spatial-compression autoencoders; (2) Decoupled High-Resolution Adaptation, an efficient decoupled three-phases training strategy for mitigating the generalization penalty of high spatial-compression autoencoders. With these designs, we improve the autoencoder's spatial compression ratio up to 128 while maintaining the reconstruction quality. Applying our DC-AE to latent diffusion models, we achieve significant speedup without accuracy drop. For example, on ImageNet 512x512, our DC-AE provides 19.1x inference speedup and 17.9x training speedup on H100 GPU for UViT-H while achieving a better FID, compared with the widely used SD-VAE-f8 autoencoder. Our code is available at //github.com/mit-han-lab/efficientvit.
Recent advancements in solving Bayesian inverse problems have spotlighted denoising diffusion models (DDMs) as effective priors. Although these have great potential, DDM priors yield complex posterior distributions that are challenging to sample. Existing approaches to posterior sampling in this context address this problem either by retraining model-specific components, leading to stiff and cumbersome methods, or by introducing approximations with uncontrolled errors that affect the accuracy of the produced samples. We present an innovative framework, divide-and-conquer posterior sampling, which leverages the inherent structure of DDMs to construct a sequence of intermediate posteriors that guide the produced samples to the target posterior. Our method significantly reduces the approximation error associated with current techniques without the need for retraining. We demonstrate the versatility and effectiveness of our approach for a wide range of Bayesian inverse problems. The code is available at \url{//github.com/Badr-MOUFAD/dcps}
The growing demand for accurate and equitable AI models in digital dermatology faces a significant challenge: the lack of diverse, high-quality labeled data. In this work, we investigate the potential of domain-specific foundation models for dermatology in addressing this challenge. We utilize self-supervised learning (SSL) techniques to pre-train models on a dataset of over 240,000 dermatological images from public and private collections. Our study considers several SSL methods and compares the resulting foundation models against domain-agnostic models like those pre-trained on ImageNet and state-of-the-art models such as MONET across 12 downstream tasks. Unlike previous research, we emphasize the development of smaller models that are more suitable for resource-limited clinical settings, facilitating easier adaptation to a broad range of use cases. Results show that models pre-trained in this work not only outperform general-purpose models but also approach the performance of models 50 times larger on clinically relevant diagnostic tasks. To promote further research in this direction, we publicly release both the training code and the foundation models, which can benefit clinicians in dermatological applications.
Effective multi-robot teams require the ability to move to goals in complex environments in order to address real-world applications such as search and rescue. Multi-robot teams should be able to operate in a completely decentralized manner, with individual robot team members being capable of acting without explicit communication between neighbors. In this paper, we propose a novel game theoretic model that enables decentralized and communication-free navigation to a goal position. Robots each play their own distributed game by estimating the behavior of their local teammates in order to identify behaviors that move them in the direction of the goal, while also avoiding obstacles and maintaining team cohesion without collisions. We prove theoretically that generated actions approach a Nash equilibrium, which also corresponds to an optimal strategy identified for each robot. We show through extensive simulations that our approach enables decentralized and communication-free navigation by a multi-robot system to a goal position, and is able to avoid obstacles and collisions, maintain connectivity, and respond robustly to sensor noise.
Learning with limited data is a key challenge for visual recognition. Few-shot learning methods address this challenge by learning an instance embedding function from seen classes and apply the function to instances from unseen classes with limited labels. This style of transfer learning is task-agnostic: the embedding function is not learned optimally discriminative with respect to the unseen classes, where discerning among them is the target task. In this paper, we propose a novel approach to adapt the embedding model to the target classification task, yielding embeddings that are task-specific and are discriminative. To this end, we employ a type of self-attention mechanism called Transformer to transform the embeddings from task-agnostic to task-specific by focusing on relating instances from the test instances to the training instances in both seen and unseen classes. Our approach also extends to both transductive and generalized few-shot classification, two important settings that have essential use cases. We verify the effectiveness of our model on two standard benchmark few-shot classification datasets --- MiniImageNet and CUB, where our approach demonstrates state-of-the-art empirical performance.
Recent advancements in deep neural networks for graph-structured data have led to state-of-the-art performance on recommender system benchmarks. However, making these methods practical and scalable to web-scale recommendation tasks with billions of items and hundreds of millions of users remains a challenge. Here we describe a large-scale deep recommendation engine that we developed and deployed at Pinterest. We develop a data-efficient Graph Convolutional Network (GCN) algorithm PinSage, which combines efficient random walks and graph convolutions to generate embeddings of nodes (i.e., items) that incorporate both graph structure as well as node feature information. Compared to prior GCN approaches, we develop a novel method based on highly efficient random walks to structure the convolutions and design a novel training strategy that relies on harder-and-harder training examples to improve robustness and convergence of the model. We also develop an efficient MapReduce model inference algorithm to generate embeddings using a trained model. We deploy PinSage at Pinterest and train it on 7.5 billion examples on a graph with 3 billion nodes representing pins and boards, and 18 billion edges. According to offline metrics, user studies and A/B tests, PinSage generates higher-quality recommendations than comparable deep learning and graph-based alternatives. To our knowledge, this is the largest application of deep graph embeddings to date and paves the way for a new generation of web-scale recommender systems based on graph convolutional architectures.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191