Knowledge distillation (KD) is the de facto standard for compressing large-scale models into smaller ones. Prior works have explored ever more complex KD strategies involving different objective functions, teacher-ensembles, and weight inheritance. In this work we explore an alternative, yet simple approach---active data curation as effective distillation for contrastive multimodal pretraining. Our simple online batch selection method, ACID, outperforms strong KD baselines across various model-, data- and compute-configurations. Further, we find such an active data curation strategy to in fact be complementary to standard KD, and can be effectively combined to train highly performant inference-efficient models. Our simple and scalable pretraining framework, ACED, achieves state-of-the-art results across 27 zero-shot classification and retrieval tasks with upto 11% less inference FLOPs. We further demonstrate that our ACED models yield strong vision-encoders for training generative multimodal models in the LiT-Decoder setting, outperforming larger vision encoders for image-captioning and visual question-answering tasks.
论文检索
输入标题、作者或关键词，从 2,871 篇学术成果中精准定位
Large pre-trained Vision-Language Models (VLMs) such as CLIP have demonstrated excellent zero-shot generalizability across various downstream tasks. However, recent studies have shown that the inference performance of CLIP can be greatly degraded by small adversarial perturbations, especially its visual modality, posing significant safety threats. To mitigate this vulnerability, in this paper, we propose a novel defense method called Test-Time Adversarial Prompt Tuning (TAPT) to enhance the inference robustness of CLIP against visual adversarial attacks. TAPT is a test-time defense method that learns defensive bimodal (textual and visual) prompts to robustify the inference process of CLIP. Specifically, it is an unsupervised method that optimizes the defensive prompts for each test sample by minimizing a multi-view entropy and aligning adversarial-clean distributions. We evaluate the effectiveness of TAPT on 11 benchmark datasets, including ImageNet and 10 other zero-shot datasets, demonstrating that it enhances the zero-shot adversarial robustness of the original CLIP by at least 48.9% against AutoAttack (AA), while largely maintaining performance on clean examples. Moreover, TAPT outperforms existing adversarial prompt tuning methods across various backbones, achieving an average robustness improvement of at least 36.6%. Code is available at https://github.com/xinwong/TAPT.
Estimating 3D motion from 2D observations is a long-standing research challenge. Prior work typically requires training on datasets containing ground truth 3D motions, limiting their applicability to activities well-represented in existing motion capture data. This dependency particularly hinders generalization to out-of-distribution scenarios or subjects where collecting 3D ground truth is challenging, such as complex athletic movements or animal motion. We introduce MVLift, a novel approach to predict global 3D motion---including both joint rotations and root trajectories in the world coordinate system---using only 2D pose sequences for training. Our multi-stage framework leverages 2D motion diffusion models to progressively generate consistent 2D pose sequences across multiple views, a key step in recovering accurate global 3D motion. MVLift generalizes across various domains, including human poses, human-object interactions, and animal poses. Despite not requiring 3D supervision, it outperforms prior work on five datasets, including those methods that require 3D supervision.
Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of the contrastive loss makes VLMs focus predominantly on foreground objects, neglecting other crucial information in the image, which limits their effectiveness in downstream tasks. To address these challenges, we propose COSMOS: CrOSs-MOdality Self-distillation for vision-language pre-training that integrates a novel text-cropping strategy and cross-attention module into a self-supervised learning framework. We create global and local views of images and texts (i.e., multi-modal augmentations), which are essential for self-distillation in VLMs. We further introduce a cross-attention module, enabling COSMOS to learn comprehensive cross-modal representations optimized via a cross-modality self-distillation loss. COSMOS consistently outperforms previous strong baselines on various zero-shot downstream tasks, including retrieval, classification, and semantic segmentation. Additionally, it surpasses CLIP-based models trained on larger datasets in visual perception and contextual understanding tasks. Code is available at https://github.com/ExplainableML/cosmos.
The development of self-supervised video-language models based on mask learning has significantly advanced downstream video tasks. These models leverage masked reconstruction to facilitate joint learning of visual and linguistic information. However, recent study reveals that reconstructing image features yields superior downstream performance compared to video feature reconstruction. We hypothesize that this performance gap stems from the way how masking strategies influence the model's attention to temporal dynamics. To validate this hypothesis, we performed two sets of experiments that demonstrate that alignment between the masked target and the reconstruction target is crucial for self-supervised video-language learning. Based on these findings, we propose a spatio-temporal masking strategy (STM) for video-language model pretraining that operates across adjacent frames, and a decoder leverages semantic information to enhance the spatio-temporal representations of masked tokens. Thanks to the combination of masking strategy and reconstruction decoder, STM enforces the model to learn spatio-temporal feature representation comprehensively. Experiments in three video understanding downstream tasks validate the superiority of our method.
Masked video modeling, such as VideoMAE, is an effective paradigm for video self-supervised learning (SSL). However, they are primarily based on reconstructing pixel level details on natural videos which have substantial temporal redundancy, limiting their capability for semantic representation and sufficient encoding of motion dynamics. To address these issues, this paper introduces a novel SSL approach for video representation learning, dubbed as SMILE, by infusing both spatial and motion semantics. In SMILE, we leverage image-language pretrained models, such as CLIP, to guide the learning process with their high-level spatial semantics. We enhance the representation of motion by introducing synthetic motion patterns in the training data, allowing the model to capture more complex and dynamic content. Furthermore, using SMILE, we establish a new self-supervised video learning paradigm capable of learning strong video representations without requiring any natural video data. We have carried out extensive experiments on 7 datasets with various downstream scenarios. SMILE surpasses current state-of-the-art SSL methods, showcasing its effectiveness in learning more discriminative and generalizable video representations.
From Zero to Detail: Deconstructing Ultra-High-Definition Image Restoration from Progressive Spectral Perspective
PDF ↗Ultra-high-definition (UHD) image restoration faces significant challenges due to its high resolution, complex content, and intricate details. To cope with these challenges, we analyze the restoration process in depth through a progressive spectral perspective, and deconstruct the complex UHD restoration problem into three progressive stages: zero-frequency enhancement, low-frequency restoration, and high-frequency refinement. Building on this insight, we propose a novel framework, ERR, which comprises three collaborative sub-networks: the zero-frequency enhancer (ZFE), the low-frequency restorer (LFR), and the high-frequency refiner (HFR). Specifically, the ZFE integrates global priors to learn global mapping, while the LFR restores low-frequency information, emphasizing reconstruction of coarse-grained content. Finally, the HFR employs our designed frequency-windowed Kolmogorov-Arnold Networks (FW-KAN) to refine textures and details, producing high-quality image restoration. Our approach significantly outperforms previous UHD methods across various tasks, with extensive ablation studies validating the effectiveness of each component.
An End-to-End Robust Point Cloud Semantic Segmentation Network with Single-Step Conditional Diffusion Models
PDF ↗Existing conditional Denoising Diffusion Probabilistic Models (DDPMs) with a Noise-Conditional Framework (NCF) remain challenging for 3D scene understanding tasks, as the complex geometric details in scenes increase the difficulty of fitting the gradients of the data distribution (the scores) from semantic labels. This also results in longer training and inference time for DDPMs compared to non-DDPMs. From a different perspective, we delve deeply into the model paradigm dominated by the Conditional Network. In this paper, we propose an end-to-end robust semantic Segmentation Network based on a Conditional-Noise Framework (CNF) of DDPMs, named CDSegNet. Specifically, CDSegNet models the Noise Network (NN) as a learnable noise-feature generator. This enables the Conditional Network (CN) to understand 3D scene semantics under multi-level feature perturbations, enhancing the generalization in unseen scenes. Meanwhile, benefiting from the noise system of DDPMs, CDSegNet exhibits strong robustness for data noise and sparsity in experiments. Moreover, thanks to CNF, CDSegNet can generate the semantic labels in a single-step inference like non-DDPMs, due to avoiding directly fitting the scores from semantic labels in the dominant network of CDSegNet. On public indoor and outdoor benchmarks, CDSegNet significantly outperforms existing methods, achieving state-of-the-art performance.
Learned Video Compression (LVC) aims to reduce redundancy in sequential data through deep learning approaches. Recent advances have significantly boosted LVC performance by shifting compression operations to feature domain, often combining Motion Estimation and Motion Compensation module(MEMC) with CNN-based context extraction. However, reliance on motions and convolution-driven context models limits generalizability and global perception. To address these issues, we propose a Feature-level Attention (FLA) module within a Transformer-based framework that perceives full-frame explicitly, thus bypassing confined motion signatures. FLA accomplishes global perception by converting high-level local patch embeddings to one-dimensional batch-wise vectors and replacing traditional attention weights to a global context matrix. Amongst this, a dense overlapping patcher (DP) is introduced to retain local features before embedding projection. Furthermore, a Transformer-CNN mixed encoder is applied to alleviate the spatial feature bottleneck without expanding latent size. Experiments demonstrate excellent generalizability with universally efficient redundancy reduction in different scenarios. Extensive tests on four video compression datasets show that our method achieves state-of-the-art Rate-Distortion performance compared to existing LVC methods and traditional codecs. A down-scaled version of our model reduced computation overhead by a great margin while maintained great performance.
Novel-view synthesis techniques achieve impressive results for static scenes but struggle when faced with the inconsistencies inherent to casual capture settings: varying illumination, scene motion, and other unintended effects that are difficult to model explicitly. We present an approach for leveraging generative video models to simulate the inconsistencies in the world that can occur during capture. We use this process, along with existing multi-view datasets, to create synthetic data for training a multi-view harmonization network that is able to reconcile inconsistent observations into a consistent 3D scene. We demonstrate that our world-simulation strategy significantly outperforms traditional augmentation methods in handling real-world scene variations, thereby enabling highly accurate static 3D reconstructions in the presence of a variety of challenging inconsistencies.
Mono-InternVL: Pushing the Boundaries of Monolithic Multimodal Large Language Models with Endogenous Visual Pre-training
PDF ↗In this paper, we focus on monolithic Multimodal Large Language Models (MLLMs) that integrate visual encoding and language decoding into a single LLM. In particular, we identify that existing pre-training strategies for monolithic MLLMs often suffer from unstable optimization or catastrophic forgetting. To address this issue, our core idea is to embed a new visual parameter space into a pre-trained LLM, thereby stably learning visual knowledge from noisy data while freezing the LLM. Based on this principle, we present Mono-InternVL, a novel monolithic MLLM that seamlessly integrates a set of visual experts via a multimodal mixture-of-experts structure. Moreover, we propose an innovative pre-training strategy to maximize the visual capability of Mono-InternVL, namely Endogenous Visual Pre-training (EViP). In particular, EViP is designed as a progressive learning process for visual experts, which aims to fully exploit the visual knowledge from noisy data to high-quality data. To validate our approach, we conduct extensive experiments on 16 benchmarks. Experimental results confirm the superior performance of Mono-InternVL over existing monolithic MLLMs on 13 of 16 multimodal benchmarks, e.g., +80 points over Emu3 on OCRBench. Compared to the modular baseline, i.e., InternVL-1.5, Mono-InternVL still retains comparable multimodal performance while reducing up to 67% first token latency. Our project is available at https://internvl.github.io/blog/2024-10-10-Mono-InternVL/.
Conventional frame-based imaging for active stereo systems has encountered major challenges in fast-motion scenarios. However, how to design a novel paradigm for high-speed depth sensing still remains an open issue. In this paper, we propose a novel problem setting, namely active event-based stereo vision, which provides the first insight of integrating binocular event cameras and an infrared projector for high-speed depth sensing. Technically, we first build a stereo camera prototype system and present a real-world dataset with over 21.5k spatiotemporal synchronized labels at 15 Hz, while also creating a realistic synthetic dataset with stereo event streams and 23.8k synchronized labels at 20 Hz. Then, we propose ActiveEventNet, a lightweight yet effective active event-based stereo matching neural network that learns to generate high-quality dense disparity maps from stereo event streams with low latency. Experiments demonstrate that our ActiveEventNet outperforms state-of-the-art methods meanwhile significantly reducing computational complexity. Our solution offers superior depth sensing compared to conventional stereo cameras in high-speed scenes, while also achieving the inference speed of up to 150 FPS with our prototype. We believe that this novel paradigm will provide new insights into future depth sensing systems. Our project can be available at https://github.com/jianing-li/active_event_based_stereo.
When a vision-language model (VLM) is prompted to identify an entity depicted in an image, it may answer "I see a conifer," rather than the specific label "Norway spruce". This raises two issues for evaluation: Firstly, the unconstrained generated text needs to be mapped to the evaluation label space (i.e., "conifer"). Secondly, a useful classification measure should give partial credit to less specific, but not incorrect, answers ("Norway spruce" being a type of "conifer"). To meet these requirements, we propose a framework for evaluating unconstrained text predictions such as those generated from a vision-language model against a taxonomy. Specifically, we propose the use of hierarchical precision and recall measures to assess the level of correctness and specificity of predictions with regard to a taxonomy. Experimentally, we first show that existing text similarity measures do not capture taxonomic similarity well. We then develop and compare different methods to map textual VLM predictions onto a taxonomy. This allows us to compute hierarchical similarity measures between the generated text and the ground truth labels. Finally, we analyze modern VLMs on fine-grained visual classification tasks based on our proposed taxonomic evaluation scheme.
SOAP: Vision-Centric 3D Semantic Scene Completion with Scene-Adaptive Decoder and Occluded Region-Aware View Projection
PDF ↗Existing view transformations in vision-centric 3D Semantic Scene Completion (SSC) inevitably experience erroneous feature duplication in the reconstructed voxel space due to occlusions, leading to a dilution of informative contexts. Furthermore, semantic classes exhibit high variability in their appearance in real-world driving scenarios. To address these issues, we introduce a novel 3D SSC method, called SOAP, including two key components: an occluded region-aware view projection and a scene-adaptive decoder. The occluded region-aware view projection effectively converts 2D image features into voxel space, refining the duplicated features of occluded regions using information gathered from previous observations. The scene-adaptive decoder guides query embeddings to learn diverse driving environments based on a comprehensive semantic repository. Extensive experiments validate that the proposed SOAP significantly outperforms existing methods for the vision-centric 3D SSC on automated driving datasets, SemanticKITTI and SSCBench. Code is available at https://github.com/gywns6287/SOAP.
The rapid advancement of multi-modal language models (MLLMs) like GPT-4o has propelled the development of Omni language models, designed to process and proactively respond to continuous streams of multi-modal data. Despite their potential, evaluating their real-world interactive capabilities in streaming video contexts remains a formidable challenge. In this work, we introduce OmniMMI, a comprehensive multi-modal interaction benchmark tailored for OmniLLMs in streaming video contexts. OmniMMI encompasses over 1,121 videos and 2,290 questions, addressing two critical yet underexplored challenges in existing video benchmarks: streaming video understanding and proactive reasoning, across six distinct subtasks. Moreover, we propose a novel framework, Multi-modal Multiplexing Modeling (M4), designed to enable an inference-efficient streaming model that can see, listen while generating.
We address the challenge of generating dynamic 4D scenes from monocular multi-object videos with heavy occlusions and introduce Robust4DGen, a novel approach that integrates rendering-based deformable 3D Gaussian optimization with generative priors for view synthesis. While existing view-synthesis models excel at novel view generation for isolated objects, they struggle with full scenes due to their complexity and data demands. To overcome this, Robust4DGen decomposes scenes into individual objects, optimizing a differentiable set of deformable Gaussians per object while capturing 2D occlusions from a 3D perspective through joint Gaussian splatting. Joint splatting ensures occlusion-aware rendering losses in observed frames while explicit object decomposition allows the usage of object-centric diffusion models for object completion in unobserved viewpoints. To reconcile the differences between object-centric priors and the global frame-centric coordinate system of the video, Robust4DGen employs differentiable transformations to unify the rendering and generative constraints within a single framework. The result is a model capable of generating 4D objects across space and time while producing 2D and 3D point tracks from monocular videos. To rigorously evaluate the quality of scene generation and the accuracy of the motion under multi-object occlusions, we introduce MOSE-PTS, a subset of the challenging MOSE benchmark, which we annotated with high-quality 2D point tracks. Quantitative evaluations and perceptual human studies confirm that Robust4DGen generates more realistic novel views of scenes and produces more accurate point tracks compared to existing approaches.
AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data
PDF ↗Recent advances in generative models have sparked research on improving model fairness with AI-generated data. However, existing methods often face limitations in the diversity and quality of synthetic data, leading to compromised fairness and overall model accuracy. Moreover, many approaches rely on the availability of demographic group labels, which are often costly to annotate. This paper proposes AIM-Fair, aiming to overcome these limitations and harness the potential of cutting-edge generative models in promoting algorithmic fairness. We investigate a fine-tuning paradigm starting from a biased model initially trained on real-world data without demographic annotations. This model is then fine-tuned using unbiased synthetic data generated by a state-of-the-art diffusion model to improve its fairness. Two key challenges are identified in this fine-tuning paradigm, 1) the low quality of synthetic data, which can still happen even with advanced generative models, and 2) the domain and bias gap between real and synthetic data. To address the limitation of synthetic data quality, we propose Contextual Synthetic Data Generation (CSDG) to generate data using a text-to-image diffusion model (T2I) with prompts generated by a context-aware LLM, ensuring both data diversity and control of bias in synthetic data. To resolve domain and bias shifts, we introduce a novel selective fine-tuning scheme in which only model parameters more sensitive to bias and less sensitive to domain shift are updated. Experiments on CelebA and UTKFace datasets show that our AIM-Fair improves model fairness while maintaining utility, outperforming both fully and partially fine-tuned approaches to model fairness.
Pansharpening aims at integrating complementary information from panchromatic and multispectral images. Available deep-learning based pansharpening methods typically perform exceptionally with particular satellite datasets. At the same time, it has been observed that these models also exhibit scene dependence, for example, if the majority of the training samples come from the urban scenes, the model's performance may decline in the river scene. To address the domain gap produced by varying satellite sensors and distinct scenes, we propose a dual-granularity semantic guided sparse routing diffusion model for general pansharpening. By utilizing the large Vision-Language Models (VLMs) in the field of geoscience, e.g, GeoChat, we introduce the dual granularity semantics to generate dynamic sparse routing scores for adaptation of different satellite sensors and scenes. This scene-level and region-level dual-granularity semantic information serves as guidance for dynamically activating specialized experts within the diffusion model. Extensive experiments on WorldView-3, QuickBird, and GaoFen-2 datasets show the effectiveness of our proposed method. Notably, the proposed method outperforms the comparison approaches in adapting to new satellite sensors and scenes. The codes are available at https://github.com/codgodtao/SGDiff.
The Base-New Trade-off (BNT) problem universally exists during the optimization of CLIP-based prompt tuning, where continuous fine-tuning on base (target) classes leads to a simultaneous decrease of generalization ability on new (unseen) classes. Existing approaches attempt to regulate the prompt tuning process to balance BNT by appending constraints. However, imposed on the same target prompt, these constraints fail to fully avert the mutual exclusivity between the optimization directions for base and new. As a novel solution to this challenge, we propose the plug-and-play Dual-Prompt Collaboration (DPC) framework, the first that decoupling the optimization processes of base and new tasks at the prompt level. Specifically, we clone a learnable parallel prompt based on the backbone prompt, and introduce a variable Weighting-Decoupling framework to independently control the optimization directions of dual prompts specific to base or new tasks, thus avoiding the conflict in generalization. Meanwhile, we propose a Dynamic Hard Negative Optimizer, utilizing dual prompts to construct a more challenging optimization task on base classes for enhancement. For interpretability, we prove the feature channel invariance of the prompt vector during the optimization process, providing theoretical support for the Weighting-Decoupling of DPC. Extensive experiments on multiple backbones demonstrate that DPC can significantly improve base performance without introducing any external knowledge beyond the base classes, while maintaining generalization to new classes. Code is available at: https://github.com/JREion/DPC.
Federated Learning (FL) is an emerging direction in distributed machine learning that enables jointly training a model without sharing the data. However, as the size of datasets grows exponentially, computational costs of FL increase. In this paper, we propose the first Coreset Selection criterion for Federated Learning (FedCS) by exploring the Distance Contrast (DC) in feature space. Our FedCS is inspired by the discovery that DC can indicate the intrinsic properties inherent to samples regardless of the networks. Based on the observation, we develop a method that is mathematically formulated to prune samples with high DC. The principle behind our pruning is that high DC samples either contain less information or represent rare extreme cases, thus removal of them can enhance the aggregation performance. Besides, we experimentally show that samples with low DC usually contain substantial information and reflect the common features of samples within their classes, such that they are suitable for constructing coreset. With only two time of linear-logarithmic complexity operation, FedCS leads to significant improvements over the methods using whole dataset in terms of computational costs, with similar accuracies. For example, on the CIFAR-10 dataset with Dirichlet coefficient \alpha=0.1, FedCS achieves 58.88% accuracy using only 44% of the entire dataset, whereas other methods require twice the data volume as FedCS for same performance.