Nanyang Technological University Singapore
Abstract:Large Language Models (LLMs) excel in diverse applications but suffer inefficiency due to massive scale. While quantization reduces computational costs, existing methods degrade accuracy in medium-sized LLMs (e.g., Llama-3-8B) due to activation outliers. To address this, we propose QUAD (Quantization with Activation Decomposition), a framework leveraging Singular Value Decomposition (SVD) to suppress activation outliers for effective 4-bit quantization. QUAD estimates activation singular vectors offline using calibration data to construct an orthogonal transformation matrix P, shifting outliers to additional dimensions in full precision while quantizing rest components to 4-bit. Additionally, QUAD enables parameter-efficient fine-tuning via adaptable full-precision outlier weights, narrowing the accuracy gap between quantized and full-precision models. Experiments demonstrate that QUAD achieves 94% ~ 96% accuracy under W4A4 quantization and 98% accuracy with W4A4/A8 and parameter-efficient fine-tuning for Llama-3 and Qwen-2.5 models. Our code is available at \href{https://github.com/hyx1999/Quad}{repository}.
Abstract:Video inpainting involves modifying local regions within a video, ensuring spatial and temporal consistency. Most existing methods focus primarily on scene completion (i.e., filling missing regions) and lack the capability to insert new objects into a scene in a controllable manner. Fortunately, recent advancements in text-to-video (T2V) diffusion models pave the way for text-guided video inpainting. However, directly adapting T2V models for inpainting remains limited in unifying completion and insertion tasks, lacks input controllability, and struggles with long videos, thereby restricting their applicability and flexibility. To address these challenges, we propose MTV-Inpaint, a unified multi-task video inpainting framework capable of handling both traditional scene completion and novel object insertion tasks. To unify these distinct tasks, we design a dual-branch spatial attention mechanism in the T2V diffusion U-Net, enabling seamless integration of scene completion and object insertion within a single framework. In addition to textual guidance, MTV-Inpaint supports multimodal control by integrating various image inpainting models through our proposed image-to-video (I2V) inpainting mode. Additionally, we propose a two-stage pipeline that combines keyframe inpainting with in-between frame propagation, enabling MTV-Inpaint to effectively handle long videos with hundreds of frames. Extensive experiments demonstrate that MTV-Inpaint achieves state-of-the-art performance in both scene completion and object insertion tasks. Furthermore, it demonstrates versatility in derived applications such as multi-modal inpainting, object editing, removal, image object brush, and the ability to handle long videos. Project page: https://mtv-inpaint.github.io/.
Abstract:Large vision-language models (LVLMs) have demonstrated remarkable achievements, yet the generation of non-factual responses remains prevalent in fact-seeking question answering (QA). Current multimodal fact-seeking benchmarks primarily focus on comparing model outputs to ground truth answers, providing limited insights into the performance of modality-specific modules. To bridge this gap, we introduce VisualSimpleQA, a multimodal fact-seeking benchmark with two key features. First, it enables streamlined and decoupled evaluation of LVLMs in visual and linguistic modalities. Second, it incorporates well-defined difficulty criteria to guide human annotation and facilitates the extraction of a challenging subset, VisualSimpleQA-hard. Experiments on 15 LVLMs show that even state-of-the-art models such as GPT-4o achieve merely 60%+ correctness in multimodal fact-seeking QA on VisualSimpleQA and 30%+ on VisualSimpleQA-hard. Furthermore, the decoupled evaluation across these models highlights substantial opportunities for improvement in both visual and linguistic modules. The dataset is available at https://huggingface.co/datasets/WYLing/VisualSimpleQA.
Abstract:Existing low-rank adaptation (LoRA) methods face challenges on sparse large language models (LLMs) due to the inability to maintain sparsity. Recent works introduced methods that maintain sparsity by augmenting LoRA techniques with additional masking mechanisms. Despite these successes, such approaches suffer from an increased memory and computation overhead, which affects efficiency of LoRA methods. In response to this limitation, we introduce LoRS, an innovative method designed to achieve both memory and computation efficiency when fine-tuning sparse LLMs. To mitigate the substantial memory and computation demands associated with preserving sparsity, our approach incorporates strategies of weight recompute and computational graph rearrangement. In addition, we also improve the effectiveness of LoRS through better adapter initialization. These innovations lead to a notable reduction in memory and computation consumption during the fine-tuning phase, all while achieving performance levels that outperform existing LoRA approaches.
Abstract:The advent of stereoscopic videos has opened new horizons in multimedia, particularly in extended reality (XR) and virtual reality (VR) applications, where immersive content captivates audiences across various platforms. Despite its growing popularity, producing stereoscopic videos remains challenging due to the technical complexities involved in generating stereo parallax. This refers to the positional differences of objects viewed from two distinct perspectives and is crucial for creating depth perception. This complex process poses significant challenges for creators aiming to deliver convincing and engaging presentations. To address these challenges, this paper introduces the Text-driven Stereoscopic Video Generation (T-SVG) system. This innovative, model-agnostic, zero-shot approach streamlines video generation by using text prompts to create reference videos. These videos are transformed into 3D point cloud sequences, which are rendered from two perspectives with subtle parallax differences, achieving a natural stereoscopic effect. T-SVG represents a significant advancement in stereoscopic content creation by integrating state-of-the-art, training-free techniques in text-to-video generation, depth estimation, and video inpainting. Its flexible architecture ensures high efficiency and user-friendliness, allowing seamless updates with newer models without retraining. By simplifying the production pipeline, T-SVG makes stereoscopic video generation accessible to a broader audience, demonstrating its potential to revolutionize the field.
Abstract:Large language models (LLMs) based on the Transformer architecture are widely employed across various domains and tasks. However, their increasing size imposes significant hardware demands, limiting practical deployment. To mitigate this, model pruning techniques have been developed to create more efficient models while maintaining high performance. Despite this, post-training after pruning is crucial for performance recovery and can be resource-intensive. This paper investigates the post-training requirements of pruned LLMs and introduces a scaling law to determine the optimal amount of post-training data. Post-training experiments with the Llama-3 and Qwen-2.5 series models, pruned using depth pruning, width pruning, and 2:4 semi-structured pruning, show that higher pruning ratios necessitate more post-training data for performance recovery, whereas larger LLMs require less. The proposed scaling law predicts a model's loss based on its parameter counts before and after pruning, as well as the post-training token counts. Furthermore, we find that the scaling law established from smaller LLMs can be reliably extrapolated to larger LLMs. This work provides valuable insights into the post-training of pruned LLMs and offers a practical scaling law for optimizing post-training data usage.
Abstract:Human motion capture is the foundation for many computer vision and graphics tasks. While industrial motion capture systems with complex camera arrays or expensive wearable sensors have been widely adopted in movie and game production, consumer-affordable and easy-to-use solutions for personal applications are still far from mature. To utilize a mixture of a monocular camera and very few inertial measurement units (IMUs) for accurate multi-modal human motion capture in daily life, we contribute MINIONS in this paper, a large-scale Motion capture dataset collected from INertial and visION Sensors. MINIONS has several featured properties: 1) large scale of over five million frames and 400 minutes duration; 2) multi-modality data of IMUs signals and RGB videos labeled with joint positions, joint rotations, SMPL parameters, etc.; 3) a diverse set of 146 fine-grained single and interactive actions with textual descriptions. With the proposed MINIONS, we conduct experiments on multi-modal motion capture and explore the possibilities of consumer-affordable motion capture using a monocular camera and very few IMUs. The experiment results emphasize the unique advantages of inertial and vision sensors, showcasing the promise of consumer-affordable multi-modal motion capture and providing a valuable resource for further research and development.
Abstract:Conventional industrial robots often use two-fingered grippers or suction cups to manipulate objects or interact with the world. Because of their simplified design, they are unable to reproduce the dexterity of human hands when manipulating a wide range of objects. While the control of humanoid hands evolved greatly, hardware platforms still lack capabilities, particularly in tactile sensing and providing soft contact surfaces. In this work, we present a method that equips the skeleton of a tendon-driven humanoid hand with a soft and sensorized tactile skin. Multi-material 3D printing allows us to iteratively approach a cast skin design which preserves the robot's dexterity in terms of range of motion and speed. We demonstrate that a soft skin enables firmer grasps and piezoresistive sensor integration enhances the hand's tactile sensing capabilities.
Abstract:Large language models (LLM) have been extensively applied in various natural language tasks and domains, but their applicability is constrained by the large number of parameters of the models. Consequently, there is an increasing emphasis on compact models that exhibit high performance. In this study, we observe that different layers in LLM have varying degrees of perturbation on the hidden states, which allows us to identify less important layers. Based on this phenomenon, we propose LLM-Streamline, which consists of two parts: layer pruning, where we remove a set of consecutive layers with the lowest importance in the model according to the target sparsity; and layer replacement, where we train a lightweight model to substitute the pruned layers, thereby mitigating the performance degradation caused by pruning. In our experiments, we utilize structures such as a multi-layer perceptron (MLP) and a transformer layer as lightweight models and ultimately demonstrate that a single MLP can effectively fit the pruned layers. Comprehensive experiments show that our proposed method, LLM-Streamline, outperforms previous state-of-the-art (SOTA) model pruning methods.
Abstract:Recent text-to-video diffusion models have achieved impressive progress. In practice, users often desire the ability to control object motion and camera movement independently for customized video creation. However, current methods lack the focus on separately controlling object motion and camera movement in a decoupled manner, which limits the controllability and flexibility of text-to-video models. In this paper, we introduce Direct-a-Video, a system that allows users to independently specify motions for one or multiple objects and/or camera movements, as if directing a video. We propose a simple yet effective strategy for the decoupled control of object motion and camera movement. Object motion is controlled through spatial cross-attention modulation using the model's inherent priors, requiring no additional optimization. For camera movement, we introduce new temporal cross-attention layers to interpret quantitative camera movement parameters. We further employ an augmentation-based approach to train these layers in a self-supervised manner on a small-scale dataset, eliminating the need for explicit motion annotation. Both components operate independently, allowing individual or combined control, and can generalize to open-domain scenarios. Extensive experiments demonstrate the superiority and effectiveness of our method. Project page: https://direct-a-video.github.io/.