Morpho-Aware Global Attention for Image Matting

Vision Transformers (ViTs) and Convolutional Neural Networks (CNNs) face inherent challenges in image matting, particularly in preserving fine structural details. ViTs, with their global receptive field enabled by the self-attention mechanism, often lose local details such as hair strands. Conversely, CNNs, constrained by their local receptive field, rely on deeper layers to approximate global context but struggle to retain fine structures at greater depths. To overcome these limitations, we propose a novel Morpho-Aware Global Attention (MAGA) mechanism, designed to effectively capture the morphology of fine structures. MAGA employs Tetris-like convolutional patterns to align the local shapes of fine structures, ensuring optimal local correspondence while maintaining sensitivity to morphological details. The extracted local morphology information is used as query embeddings, which are projected onto global key embeddings to emphasize local details in a broader context. Subsequently, by projecting onto value embeddings, MAGA seamlessly integrates these emphasized morphological details into a unified global structure. This approach enables MAGA to simultaneously focus on local morphology and unify these details into a coherent whole, effectively preserving fine structures. Extensive experiments show that our MAGA-based ViT achieves significant performance gains, outperforming state-of-the-art methods across two benchmarks with average improvements of 4.3% in SAD and 39.5% in MSE.

NEO: Saving GPU Memory Crisis with CPU Offloading for Online LLM Inference

Online LLM inference powers many exciting applications such as intelligent chatbots and autonomous agents. Modern LLM inference engines widely rely on request batching to improve inference throughput, aiming to make it cost-efficient when running on expensive GPU accelerators. However, the limited GPU memory has largely limited the batch size achieved in practice, leaving significant GPU compute resources wasted. We present NEO, an online LLM inference system that offloads part of attention compute and KV cache states from the GPU to the local host CPU, effectively increasing the GPU batch size and thus inference throughput. To this end, NEO proposes asymmetric GPU-CPU pipelining and load-aware scheduling to balance GPU and CPU loads and fully utilize their compute and memory resources. We evaluate NEO on a wide range of workloads (i.e., code generation, text summarization), GPUs (i.e., T4, A10G, H100), and LLM models (i.e., 7B, 8B, 70B). NEO achieves up to 7.5$\times$, 26%, and 14% higher throughput compared to GPU-only approach on T4, A10G, and H100 GPUs, respectively, while maintaining the same latency; with more powerful CPUs, NEO achieves up to 79.3% throughput gain on A10G GPU.

BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays

Medical Vision-Language Pretraining (MedVLP) shows promise in learning generalizable and transferable visual representations from paired and unpaired medical images and reports. MedVLP can provide useful features to downstream tasks and facilitate adapting task-specific models to new setups using fewer examples. However, existing MedVLP methods often differ in terms of datasets, preprocessing, and finetuning implementations. This pose great challenges in evaluating how well a MedVLP method generalizes to various clinically-relevant tasks due to the lack of unified, standardized, and comprehensive benchmark. To fill this gap, we propose BenchX, a unified benchmark framework that enables head-to-head comparison and systematical analysis between MedVLP methods using public chest X-ray datasets. Specifically, BenchX is composed of three components: 1) Comprehensive datasets covering nine datasets and four medical tasks; 2) Benchmark suites to standardize data preprocessing, train-test splits, and parameter selection; 3) Unified finetuning protocols that accommodate heterogeneous MedVLP methods for consistent task adaptation in classification, segmentation, and report generation, respectively. Utilizing BenchX, we establish baselines for nine state-of-the-art MedVLP methods and found that the performance of some early MedVLP methods can be enhanced to surpass more recent ones, prompting a revisiting of the developments and conclusions from prior works in MedVLP. Our code are available at https://github.com/yangzhou12/BenchX.

DQRM: Deep Quantized Recommendation Models

Large-scale recommendation models are currently the dominant workload for many large Internet companies. These recommenders are characterized by massive embedding tables that are sparsely accessed by the index for user and item features. The size of these 1TB+ tables imposes a severe memory bottleneck for the training and inference of recommendation models. In this work, we propose a novel recommendation framework that is small, powerful, and efficient to run and train, based on the state-of-the-art Deep Learning Recommendation Model (DLRM). The proposed framework makes inference more efficient on the cloud servers, explores the possibility of deploying powerful recommenders on smaller edge devices, and optimizes the workload of the communication overhead in distributed training under the data parallelism settings. Specifically, we show that quantization-aware training (QAT) can impose a strong regularization effect to mitigate the severe overfitting issues suffered by DLRMs. Consequently, we achieved INT4 quantization of DLRM models without any accuracy drop. We further propose two techniques that improve and accelerate the conventional QAT workload specifically for the embedding tables in the recommendation models. Furthermore, to achieve efficient training, we quantize the gradients of the embedding tables into INT8 on top of the well-supported specified sparsification. We show that combining gradient sparsification and quantization together significantly reduces the amount of communication. Briefly, DQRM models with INT4 can achieve 79.07% accuracy on Kaggle with 0.27 GB model size, and 81.21% accuracy on the Terabyte dataset with 1.57 GB, which even outperform FP32 DLRMs that have much larger model sizes (2.16 GB on Kaggle and 12.58 on Terabyte).

AttriPrompter: Auto-Prompting with Attribute Semantics for Zero-shot Nuclei Detection via Visual-Language Pre-trained Models

Large-scale visual-language pre-trained models (VLPMs) have demonstrated exceptional performance in downstream object detection through text prompts for natural scenes. However, their application to zero-shot nuclei detection on histopathology images remains relatively unexplored, mainly due to the significant gap between the characteristics of medical images and the web-originated text-image pairs used for pre-training. This paper aims to investigate the potential of the object-level VLPM, Grounded Language-Image Pre-training (GLIP), for zero-shot nuclei detection. Specifically, we propose an innovative auto-prompting pipeline, named AttriPrompter, comprising attribute generation, attribute augmentation, and relevance sorting, to avoid subjective manual prompt design. AttriPrompter utilizes VLPMs' text-to-image alignment to create semantically rich text prompts, which are then fed into GLIP for initial zero-shot nuclei detection. Additionally, we propose a self-trained knowledge distillation framework, where GLIP serves as the teacher with its initial predictions used as pseudo labels, to address the challenges posed by high nuclei density, including missed detections, false positives, and overlapping instances. Our method exhibits remarkable performance in label-free nuclei detection, outperforming all existing unsupervised methods and demonstrating excellent generality. Notably, this work highlights the astonishing potential of VLPMs pre-trained on natural image-text pairs for downstream tasks in the medical field as well. Code will be released at https://github.com/wuyongjianCODE/AttriPrompter.

MagicPIG: LSH Sampling for Efficient LLM Generation

Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by $1.9\sim3.9\times$ across various GPU hardware and achieve 110ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens. The code is available at \url{https://github.com/Infini-AI-Lab/MagicPIG}.

The Curse of Multi-Modalities: Evaluating Hallucinations of Large Multimodal Models across Language, Visual, and Audio

Recent advancements in large multimodal models (LMMs) have significantly enhanced performance across diverse tasks, with ongoing efforts to further integrate additional modalities such as video and audio. However, most existing LMMs remain vulnerable to hallucinations, the discrepancy between the factual multimodal input and the generated textual output, which has limited their applicability in various real-world scenarios. This paper presents the first systematic investigation of hallucinations in LMMs involving the three most common modalities: language, visual, and audio. Our study reveals two key contributors to hallucinations: overreliance on unimodal priors and spurious inter-modality correlations. To address these challenges, we introduce the benchmark The Curse of Multi-Modalities (CMM), which comprehensively evaluates hallucinations in LMMs, providing a detailed analysis of their underlying issues. Our findings highlight key vulnerabilities, including imbalances in modality integration and biases from training data, underscoring the need for balanced cross-modal learning and enhanced hallucination mitigation strategies. Based on our observations and findings, we suggest potential research directions that could enhance the reliability of LMMs.

SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction

Predicting the future motion of surrounding agents is essential for autonomous vehicles (AVs) to operate safely in dynamic, human-robot-mixed environments. However, the scarcity of large-scale driving datasets has hindered the development of robust and generalizable motion prediction models, limiting their ability to capture complex interactions and road geometries. Inspired by recent advances in natural language processing (NLP) and computer vision (CV), self-supervised learning (SSL) has gained significant attention in the motion prediction community for learning rich and transferable scene representations. Nonetheless, existing pre-training methods for motion prediction have largely focused on specific model architectures and single dataset, limiting their scalability and generalizability. To address these challenges, we propose SmartPretrain, a general and scalable SSL framework for motion prediction that is both model-agnostic and dataset-agnostic. Our approach integrates contrastive and reconstructive SSL, leveraging the strengths of both generative and discriminative paradigms to effectively represent spatiotemporal evolution and interactions without imposing architectural constraints. Additionally, SmartPretrain employs a dataset-agnostic scenario sampling strategy that integrates multiple datasets, enhancing data volume, diversity, and robustness. Extensive experiments on multiple datasets demonstrate that SmartPretrain consistently improves the performance of state-of-the-art prediction models across datasets, data splits and main metrics. For instance, SmartPretrain significantly reduces the MissRate of Forecast-MAE by 10.6%. These results highlight SmartPretrain's effectiveness as a unified, scalable solution for motion prediction, breaking free from the limitations of the small-data regime. Codes are available at https://github.com/youngzhou1999/SmartPretrain

Progressive Autoregressive Video Diffusion Models

Current frontier video diffusion models have demonstrated remarkable results at generating high-quality videos. However, they can only generate short video clips, normally around 10 seconds or 240 frames, due to computation limitations during training. In this work, we show that existing models can be naturally extended to autoregressive video diffusion models without changing the architectures. Our key idea is to assign the latent frames with progressively increasing noise levels rather than a single noise level, which allows for fine-grained condition among the latents and large overlaps between the attention windows. Such progressive video denoising allows our models to autoregressively generate video frames without quality degradation or abrupt scene changes. We present state-of-the-art results on long video generation at 1 minute (1440 frames at 24 FPS). Videos from this paper are available at https://desaixie.github.io/pa-vdm/.

HARIVO: Harnessing Text-to-Image Models for Video Generation

We present a method to create diffusion-based video models from pretrained Text-to-Image (T2I) models. Recently, AnimateDiff proposed freezing the T2I model while only training temporal layers. We advance this method by proposing a unique architecture, incorporating a mapping network and frame-wise tokens, tailored for video generation while maintaining the diversity and creativity of the original T2I model. Key innovations include novel loss functions for temporal smoothness and a mitigating gradient sampling technique, ensuring realistic and temporally consistent video generation despite limited public video data. We have successfully integrated video-specific inductive biases into the architecture and loss functions. Our method, built on the frozen StableDiffusion model, simplifies training processes and allows for seamless integration with off-the-shelf models like ControlNet and DreamBooth. project page: https://kwonminki.github.io/HARIVO