Large foundation models, including large language models (LLMs), vision transformers (ViTs), diffusion, and LLM-based multimodal models, are revolutionizing the entire machine learning lifecycle, from training to deployment. However, the substantial advancements in versatility and performance these models offer come at a significant cost in terms of hardware resources. To support the growth of these large models in a scalable and environmentally sustainable way, there has been a considerable focus on developing resource-efficient strategies. This survey delves into the critical importance of such research, examining both algorithmic and systemic aspects. It offers a comprehensive analysis and valuable insights gleaned from existing literature, encompassing a broad array of topics from cutting-edge model architectures and training/serving algorithms to practical system designs and implementations. The goal of this survey is to provide an overarching understanding of how current approaches are tackling the resource challenges posed by large foundation models and to potentially inspire future breakthroughs in this field.
Traditional channel-wise pruning methods by reducing network channels struggle to effectively prune efficient CNN models with depth-wise convolutional layers and certain efficient modules, such as popular inverted residual blocks. Prior depth pruning methods by reducing network depths are not suitable for pruning some efficient models due to the existence of some normalization layers. Moreover, finetuning subnet by directly removing activation layers would corrupt the original model weights, hindering the pruned model from achieving high performance. To address these issues, we propose a novel depth pruning method for efficient models. Our approach proposes a novel block pruning strategy and progressive training method for the subnet. Additionally, we extend our pruning method to vision transformer models. Experimental results demonstrate that our method consistently outperforms existing depth pruning methods across various pruning configurations. We obtained three pruned ConvNeXtV1 models with our method applying on ConvNeXtV1, which surpass most SOTA efficient models with comparable inference performance. Our method also achieves state-of-the-art pruning performance on the vision transformer model.
Adaptive video streaming requires efficient bitrate ladder construction to meet heterogeneous network conditions and end-user demands. Per-title optimized encoding typically traverses numerous encoding parameters to search the Pareto-optimal operating points for each video. Recently, researchers have attempted to predict the content-optimized bitrate ladder for pre-encoding overhead reduction. However, existing methods commonly estimate the encoding parameters on the Pareto front and still require subsequent pre-encodings. In this paper, we propose to directly predict the optimal transcoding resolution at each preset bitrate for efficient bitrate ladder construction. We adopt a Temporal Attentive Gated Recurrent Network to capture spatial-temporal features and predict transcoding resolutions as a multi-task classification problem. We demonstrate that content-optimized bitrate ladders can thus be efficiently determined without any pre-encoding. Our method well approximates the ground-truth bitrate-resolution pairs with a slight Bj{\o}ntegaard Delta rate loss of 1.21% and significantly outperforms the state-of-the-art fixed ladder.
Forecasting future stock trends remains challenging for academia and industry due to stochastic inter-stock dynamics and hierarchical intra-stock dynamics influencing stock prices. In recent years, graph neural networks have achieved remarkable performance in this problem by formulating multiple stocks as graph-structured data. However, most of these approaches rely on artificially defined factors to construct static stock graphs, which fail to capture the intrinsic interdependencies between stocks that rapidly evolve. In addition, these methods often ignore the hierarchical features of the stocks and lose distinctive information within. In this work, we propose a novel graph learning approach implemented without expert knowledge to address these issues. First, our approach automatically constructs dynamic stock graphs by entropy-driven edge generation from a signal processing perspective. Then, we further learn task-optimal dependencies between stocks via a generalized graph diffusion process on constructed stock graphs. Last, a decoupled representation learning scheme is adopted to capture distinctive hierarchical intra-stock features. Experimental results demonstrate substantial improvements over state-of-the-art baselines on real-world datasets. Moreover, the ablation study and sensitivity study further illustrate the effectiveness of the proposed method in modeling the time-evolving inter-stock and intra-stock dynamics.
Crowd counting has gained significant popularity due to its practical applications. However, mainstream counting methods ignore precise individual localization and suffer from annotation noise because of counting from estimating density maps. Additionally, they also struggle with high-density images.To address these issues, we propose an end-to-end model called Fine-Grained Extraction Network (FGENet). Different from methods estimating density maps, FGENet directly learns the original coordinate points that represent the precise localization of individuals.This study designs a fusion module, named Fine-Grained Feature Pyramid(FGFP), that is used to fuse feature maps extracted by the backbone of FGENet. The fused features are then passed to both regression and classification heads, where the former provides predicted point coordinates for a given image, and the latter determines the confidence level for each predicted point being an individual. At the end, FGENet establishes correspondences between prediction points and ground truth points by employing the Hungarian algorithm. For training FGENet, we design a robust loss function, named Three-Task Combination (TTC), to mitigate the impact of annotation noise. Extensive experiments are conducted on four widely used crowd counting datasets. Experimental results demonstrate the effectiveness of FGENet. Notably, our method achieves a remarkable improvement of 3.14 points in Mean Absolute Error (MAE) on the ShanghaiTech Part A dataset, showcasing its superiority over the existing state-of-the-art methods. Even more impressively, FGENet surpasses previous benchmarks on the UCF\_CC\_50 dataset with an astounding enhancement of 30.16 points in MAE.
The issue of generative pretraining for vision models has persisted as a long-standing conundrum. At present, the text-to-image (T2I) diffusion model demonstrates remarkable proficiency in generating high-definition images matching textual inputs, a feat made possible through its pre-training on large-scale image-text pairs. This leads to a natural inquiry: can diffusion models be utilized to tackle visual perception tasks? In this paper, we propose a simple yet effective scheme to harness a diffusion model for visual perception tasks. Our key insight is to introduce learnable embeddings (meta prompts) to the pre-trained diffusion models to extract proper features for perception. The effect of meta prompts are two-fold. First, as a direct replacement of the text embeddings in the T2I models, it can activate task-relevant features during feature extraction. Second, it will be used to re-arrange the extracted features to ensures that the model focuses on the most pertinent features for the task on hand. Additionally, we design a recurrent refinement training strategy that fully leverages the property of diffusion models, thereby yielding stronger visual features. Extensive experiments across various benchmarks validate the effectiveness of our approach. Our approach achieves new performance records in depth estimation tasks on NYU depth V2 and KITTI, and in semantic segmentation task on CityScapes. Concurrently, the proposed method attains results comparable to the current state-of-the-art in semantic segmentation on ADE20K and pose estimation on COCO datasets, further exemplifying its robustness and versatility.
The scarcity of labeled audio-visual datasets is a constraint for training superior audio-visual speaker diarization systems. To improve the performance of audio-visual speaker diarization, we leverage pre-trained supervised and self-supervised speech models for audio-visual speaker diarization. Specifically, we adopt supervised~(ResNet and ECAPA-TDNN) and self-supervised pre-trained models~(WavLM and HuBERT) as the speaker and audio embedding extractors in an end-to-end audio-visual speaker diarization~(AVSD) system. Then we explore the effectiveness of different frameworks, including Transformer, Conformer, and cross-attention mechanism, in the audio-visual decoder. To mitigate the degradation of performance caused by separate training, we jointly train the audio encoder, speaker encoder, and audio-visual decoder in the AVSD system. Experiments on the MISP dataset demonstrate that the proposed method achieves superior performance and obtained third place in MISP Challenge 2022.
Generating multi-camera street-view videos is critical for augmenting autonomous driving datasets, addressing the urgent demand for extensive and varied data. Due to the limitations in diversity and challenges in handling lighting conditions, traditional rendering-based methods are increasingly being supplanted by diffusion-based methods. However, a significant challenge in diffusion-based methods is ensuring that the generated sensor data preserve both intra-world consistency and inter-sensor coherence. To address these challenges, we combine an additional explicit world volume and propose the World Volume-aware Multi-camera Driving Scene Generator (WoVoGen). This system is specifically designed to leverage 4D world volume as a foundational element for video generation. Our model operates in two distinct phases: (i) envisioning the future 4D temporal world volume based on vehicle control sequences, and (ii) generating multi-camera videos, informed by this envisioned 4D temporal world volume and sensor interconnectivity. The incorporation of the 4D world volume empowers WoVoGen not only to generate high-quality street-view videos in response to vehicle control inputs but also to facilitate scene editing tasks.
Large vision-language models (VLMs) have garnered increasing interest in autonomous driving areas, due to their advanced capabilities in complex reasoning tasks essential for highly autonomous vehicle behavior. Despite their potential, research in autonomous systems is hindered by the lack of datasets with annotated reasoning chains that explain the decision-making processes in driving. To bridge this gap, we present Reason2Drive, a benchmark dataset with over 600K video-text pairs, aimed at facilitating the study of interpretable reasoning in complex driving environments. We distinctly characterize the autonomous driving process as a sequential combination of perception, prediction, and reasoning steps, and the question-answer pairs are automatically collected from a diverse range of open-source outdoor driving datasets, including nuScenes, Waymo and ONCE. Moreover, we introduce a novel aggregated evaluation metric to assess chain-based reasoning performance in autonomous systems, addressing the semantic ambiguities of existing metrics such as BLEU and CIDEr. Based on the proposed benchmark, we conduct experiments to assess various existing VLMs, revealing insights into their reasoning capabilities. Additionally, we develop an efficient approach to empower VLMs to leverage object-level perceptual elements in both feature extraction and prediction, further enhancing their reasoning accuracy. The code and dataset will be released.
Modeling dynamic, large-scale urban scenes is challenging due to their highly intricate geometric structures and unconstrained dynamics in both space and time. Prior methods often employ high-level architectural priors, separating static and dynamic elements, resulting in suboptimal capture of their synergistic interactions. To address this challenge, we present a unified representation model, called Periodic Vibration Gaussian (PVG). PVG builds upon the efficient 3D Gaussian splatting technique, originally designed for static scene representation, by introducing periodic vibration-based temporal dynamics. This innovation enables PVG to elegantly and uniformly represent the characteristics of various objects and elements in dynamic urban scenes. To enhance temporally coherent representation learning with sparse training data, we introduce a novel flow-based temporal smoothing mechanism and a position-aware adaptive control strategy. Extensive experiments on Waymo Open Dataset and KITTI benchmarks demonstrate that PVG surpasses state-of-the-art alternatives in both reconstruction and novel view synthesis for both dynamic and static scenes. Notably, PVG achieves this without relying on manually labeled object bounding boxes or expensive optical flow estimation. Moreover, PVG exhibits 50/6000-fold acceleration in training/rendering over the best alternative.