Learning a robust vision model despite large distribution shift is essential for model deployment in real-world settings. Especially, domain generalization (DG) algorithm aims to maintain the performance of a trained model on different distributions which were not seen during training. One of the most effective methods has been leveraging the already learned rich knowledge of large pretrained models. However, naively fine-tuning large models to DG tasks is often practically infeasible due to memory limitations, extensive time requirements for training, and the risk of learned knowledge deterioration. Recently, parameter-efficient fine-tuning (PEFT) methods have been proposed to reduce the high computational cost during training and efficiently adapt large models to downstream tasks. In this work, for the first time, we find that the use of adapters in PEFT methods not only reduce high computational cost during training but also serve as an effective regularizer for DG tasks. Surprisingly, a naive adapter implementation for large models achieve superior performance on common datasets. However, in situations of large distribution shifts, additional factors such as optimal amount of regularization due to the strength of distribution shifts should be considered for a sophisticated adapter implementation. To address this, we propose a mixture-of-expert based adapter fine-tuning method, dubbed as mixture-of-adapters (MoA). Specifically, we employ multiple adapters that have varying capacities, and by using learnable routers, we allocate each token to a proper adapter. By using both PEFT and MoA methods, we effectively alleviate the performance deterioration caused by distribution shifts and achieve state-of-the-art performance on diverse DG benchmarks.
相關內容
Monocular depth estimation is a fundamental computer vision task. Recovering 3D depth from a single image is geometrically ill-posed and requires scene understanding, so it is not surprising that the rise of deep learning has led to a breakthrough. The impressive progress of monocular depth estimators has mirrored the growth in model capacity, from relatively modest CNNs to large Transformer architectures. Still, monocular depth estimators tend to struggle when presented with images with unfamiliar content and layout, since their knowledge of the visual world is restricted by the data seen during training, and challenged by zero-shot generalization to new domains. This motivates us to explore whether the extensive priors captured in recent generative diffusion models can enable better, more generalizable depth estimation. We introduce Marigold, a method for affine-invariant monocular depth estimation that is derived from Stable Diffusion and retains its rich prior knowledge. The estimator can be fine-tuned in a couple of days on a single GPU using only synthetic training data. It delivers state-of-the-art performance across a wide range of datasets, including over 20% performance gains in specific cases. Project page: //marigoldmonodepth.github.io.
Global visual localization estimates the absolute pose of a camera using a single image, in a previously mapped area. Obtaining the pose from a single image enables many robotics and augmented/virtual reality applications. Inspired by latest advances in deep learning, many existing approaches directly learn and regress 6 DoF pose from an input image. However, these methods do not fully utilize the underlying scene geometry for pose regression. The challenge in monocular relocalization is the minimal availability of supervised training data, which is just the corresponding 6 DoF poses of the images. In this paper, we propose to utilize these minimal available labels (.i.e, poses) to learn the underlying 3D geometry of the scene and use the geometry to estimate the 6 DoF camera pose. We present a learning method that uses these pose labels and rigid alignment to learn two 3D geometric representations (\textit{X, Y, Z coordinates}) of the scene, one in camera coordinate frame and the other in global coordinate frame. Given a single image, it estimates these two 3D scene representations, which are then aligned to estimate a pose that matches the pose label. This formulation allows for the active inclusion of additional learning constraints to minimize 3D alignment errors between the two 3D scene representations, and 2D re-projection errors between the 3D global scene representation and 2D image pixels, resulting in improved localization accuracy. During inference, our model estimates the 3D scene geometry in camera and global frames and aligns them rigidly to obtain pose in real-time. We evaluate our work on three common visual localization datasets, conduct ablation studies, and show that our method exceeds state-of-the-art regression methods' pose accuracy on all datasets.
The goal of weakly supervised video anomaly detection is to learn a detection model using only video-level labeled data. However, prior studies typically divide videos into fixed-length segments without considering the complexity or duration of anomalies. Moreover, these studies usually just detect the most abnormal segments, potentially overlooking the completeness of anomalies. To address these limitations, we propose a Dynamic Erasing Network (DE-Net) for weakly supervised video anomaly detection, which learns multi-scale temporal features. Specifically, to handle duration variations of abnormal events, we first propose a multi-scale temporal modeling module, capable of extracting features from segments of varying lengths and capturing both local and global visual information across different temporal scales. Then, we design a dynamic erasing strategy, which dynamically assesses the completeness of the detected anomalies and erases prominent abnormal segments in order to encourage the model to discover gentle abnormal segments in a video. The proposed method obtains favorable performance compared to several state-of-the-art approaches on three datasets: XD-Violence, TAD, and UCF-Crime. Code will be made available at //github.com/ArielZc/DE-Net.
The introduction of assistive construction robots can significantly alleviate physical demands on construction workers. Leveraging a Building Information Model (BIM) offers a natural and promising approach to driving a robotic construction workflow. However, because of uncertainties inherent in construction sites, such as discrepancies between the as-designed and as-built components, robots cannot solely rely on a BIM to plan and perform field construction work. Human workers are adept at improvising alternative plans with their creativity and experience and thus can assist robots in overcoming uncertainties and performing construction work successfully. In such scenarios, it is critical to continuously update the BIM as work processes unfold so that it includes as-built information for the ensuing construction and maintenance tasks. This research introduces an interactive closed-loop digital twin framework that integrates a BIM into human-robot collaborative construction workflows. The robot's functions are primarily driven by the BIM, but it adaptively adjusts its plans based on actual site conditions, while the human co-worker oversees and supervises the process. When necessary, the human co-worker intervenes in the robot's plan by changing the task sequence or workspace geometry or requesting a new motion plan to help the robot overcome the encountered uncertainties. Experiments involving block pick-and-place tasks are carried out to verify system performance using an industrial robotic arm in a research laboratory setting that mimics a construction site. In addition, a drywall installation case study is conducted to validate the system. Integrating the flexibility of human workers and the autonomy and accuracy afforded by BIMs, the proposed framework offers significant promise of increasing the robustness of construction robots in the performance of field construction work.
There is a growing demand for explainable, transparent, and data-driven models within the domain of fraud detection. Decisions made by fraud detection models need to be explainable in the event of a customer dispute. Additionally, the decision-making process in the model must be transparent to win the trust of regulators and business stakeholders. At the same time, fraud detection solutions can benefit from data due to the noisy, dynamic nature of fraud and the availability of large historical data sets. Finally, fraud detection is notorious for its class imbalance: there are typically several orders of magnitude more legitimate transactions than fraudulent ones. In this paper, we present Deep Symbolic Classification (DSC), an extension of the Deep Symbolic Regression framework to classification problems. DSC casts classification as a search problem in the space of all analytic functions composed of a vocabulary of variables, constants, and operations and optimizes for an arbitrary evaluation metric directly. The search is guided by a deep neural network trained with reinforcement learning. Because the functions are mathematical expressions that are in closed-form and concise, the model is inherently explainable both at the level of a single classification decision and the model's decision process. Furthermore, the class imbalance problem is successfully addressed by optimizing for metrics that are robust to class imbalance such as the F1 score. This eliminates the need for oversampling and undersampling techniques that plague traditional approaches. Finally, the model allows to explicitly balance between the prediction accuracy and the explainability. An evaluation on the PaySim data set demonstrates competitive predictive performance with state-of-the-art models, while surpassing them in terms of explainability. This establishes DSC as a promising model for fraud detection systems.
We present MicroCinema, a straightforward yet effective framework for high-quality and coherent text-to-video generation. Unlike existing approaches that align text prompts with video directly, MicroCinema introduces a Divide-and-Conquer strategy which divides the text-to-video into a two-stage process: text-to-image generation and image\&text-to-video generation. This strategy offers two significant advantages. a) It allows us to take full advantage of the recent advances in text-to-image models, such as Stable Diffusion, Midjourney, and DALLE, to generate photorealistic and highly detailed images. b) Leveraging the generated image, the model can allocate less focus to fine-grained appearance details, prioritizing the efficient learning of motion dynamics. To implement this strategy effectively, we introduce two core designs. First, we propose the Appearance Injection Network, enhancing the preservation of the appearance of the given image. Second, we introduce the Appearance Noise Prior, a novel mechanism aimed at maintaining the capabilities of pre-trained 2D diffusion models. These design elements empower MicroCinema to generate high-quality videos with precise motion, guided by the provided text prompts. Extensive experiments demonstrate the superiority of the proposed framework. Concretely, MicroCinema achieves SOTA zero-shot FVD of 342.86 on UCF-101 and 377.40 on MSR-VTT. See //wangyanhui666.github.io/MicroCinema.github.io/ for video samples.
Transformer-based models have greatly pushed the boundaries of time series forecasting recently. Existing methods typically encode time series data into $\textit{patches}$ using one or a fixed set of patch lengths. This, however, could result in a lack of ability to capture the variety of intricate temporal dependencies present in real-world multi-periodic time series. In this paper, we propose MultiResFormer, which dynamically models temporal variations by adaptively choosing optimal patch lengths. Concretely, at the beginning of each layer, time series data is encoded into several parallel branches, each using a detected periodicity, before going through the transformer encoder block. We conduct extensive evaluations on long- and short-term forecasting datasets comparing MultiResFormer with state-of-the-art baselines. MultiResFormer outperforms patch-based Transformer baselines on long-term forecasting tasks and also consistently outperforms CNN baselines by a large margin, while using much fewer parameters than these baselines.
This study focuses on a novel task in text-to-image (T2I) generation, namely action customization. The objective of this task is to learn the co-existing action from limited data and generalize it to unseen humans or even animals. Experimental results show that existing subject-driven customization methods fail to learn the representative characteristics of actions and struggle in decoupling actions from context features, including appearance. To overcome the preference for low-level features and the entanglement of high-level features, we propose an inversion-based method Action-Disentangled Identifier (ADI) to learn action-specific identifiers from the exemplar images. ADI first expands the semantic conditioning space by introducing layer-wise identifier tokens, thereby increasing the representational richness while distributing the inversion across different features. Then, to block the inversion of action-agnostic features, ADI extracts the gradient invariance from the constructed sample triples and masks the updates of irrelevant channels. To comprehensively evaluate the task, we present an ActionBench that includes a variety of actions, each accompanied by meticulously selected samples. Both quantitative and qualitative results show that our ADI outperforms existing baselines in action-customized T2I generation. Our project page is at //adi-t2i.github.io/ADI.
Existing knowledge graph (KG) embedding models have primarily focused on static KGs. However, real-world KGs do not remain static, but rather evolve and grow in tandem with the development of KG applications. Consequently, new facts and previously unseen entities and relations continually emerge, necessitating an embedding model that can quickly learn and transfer new knowledge through growth. Motivated by this, we delve into an expanding field of KG embedding in this paper, i.e., lifelong KG embedding. We consider knowledge transfer and retention of the learning on growing snapshots of a KG without having to learn embeddings from scratch. The proposed model includes a masked KG autoencoder for embedding learning and update, with an embedding transfer strategy to inject the learned knowledge into the new entity and relation embeddings, and an embedding regularization method to avoid catastrophic forgetting. To investigate the impacts of different aspects of KG growth, we construct four datasets to evaluate the performance of lifelong KG embedding. Experimental results show that the proposed model outperforms the state-of-the-art inductive and lifelong embedding baselines.
Translational distance-based knowledge graph embedding has shown progressive improvements on the link prediction task, from TransE to the latest state-of-the-art RotatE. However, N-1, 1-N and N-N predictions still remain challenging. In this work, we propose a novel translational distance-based approach for knowledge graph link prediction. The proposed method includes two-folds, first we extend the RotatE from 2D complex domain to high dimension space with orthogonal transforms to model relations for better modeling capacity. Second, the graph context is explicitly modeled via two directed context representations. These context representations are used as part of the distance scoring function to measure the plausibility of the triples during training and inference. The proposed approach effectively improves prediction accuracy on the difficult N-1, 1-N and N-N cases for knowledge graph link prediction task. The experimental results show that it achieves better performance on two benchmark data sets compared to the baseline RotatE, especially on data set (FB15k-237) with many high in-degree connection nodes.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191