Composed Image Retrieval (CIR) aims to retrieve a target image based on a query composed of a reference image and a relative caption that describes the difference between the two images. The high effort and cost required for labeling datasets for CIR hamper the widespread usage of existing methods, as they rely on supervised learning. In this work, we propose a new task, Zero-Shot CIR (ZS-CIR), that aims to address CIR without requiring a labeled training dataset. Our approach, named zero-Shot composEd imAge Retrieval with textuaL invErsion SEARLE, maps the visual features of the reference image into a pseudo-word token in CLIP token embedding space and integrates it with the relative caption. To support research on ZS-CIR, we introduce an open-domain benchmarking dataset named Composed Image Retrieval on Common Objects in context (CIRCO), which is the first dataset for CIR containing multiple ground-truths for each query. The experiments show that SEARLE exhibits better performance than the baselines on the two main datasets for CIR tasks, FashionIQ and CIRR, and on the proposed CIRCO. The dataset, the code and the model are publicly available at https://github.com/miccunifi/SEARLE.
论文检索
输入标题、作者或关键词，从 8,216 篇学术成果中精准定位
SVDFormer: Complementing Point Cloud via Self-view Augmentation and Self-structure Dual-generator
PDF ↗In this paper, we propose a novel network, SVDFormer, to tackle two specific challenges in point cloud completion: understanding faithful global shapes from incomplete point clouds and generating high-accuracy local structures. Current methods either perceive shape patterns using only 3D coordinates or import extra images with well-calibrated intrinsic parameters to guide the geometry estimation of the missing parts. However, these approaches do not always fully leverage the cross-modal self-structures available for accurate and high-quality point cloud completion. To this end, we first design a Self-view Fusion Network that leverages multiple-view depth image information to observe incomplete self-shape and generate a compact global shape. To reveal highly detailed structures, we then introduce a refinement module, called Self-structure Dual-generator, in which we incorporate learned shape priors and geometric self-similarities for producing new points. By perceiving the incompleteness of each point, the dual-path design disentangles refinement strategies conditioned on the structural type of each point. SVDFormer absorbs the wisdom of self-structures, avoiding any additional paired information such as color images with precisely calibrated camera intrinsic parameters. Comprehensive experiments indicate that our method achieves state-of-the-art performance on widely-used benchmarks. Code is available at https://github.com/czvvd/SVDFormer.
In this paper, we develop rotation-equivariant neural networks for 4D panoptic segmentation. 4D panoptic segmentation is a benchmark task for autonomous driving that requires recognizing semantic classes and object instances on the road based on LiDAR scans, as well as assigning temporally consistent IDs to instances across time. We observe that the driving scenario is symmetric to rotations on the ground plane. Therefore, rotation-equivariance could provide better generalization and more robust feature learning. Specifically, we review the object instance clustering strategies and restate the centerness-based approach and the offset-based approach as the prediction of invariant scalar fields and equivariant vector fields. Other sub-tasks are also unified from this perspective, and different invariant and equivariant layers are designed to facilitate their predictions. Through evaluation on the standard 4D panoptic segmentation benchmark of SemanticKITTI, we show that our equivariant models achieve higher accuracy with lower computational costs compared to their non-equivariant counterparts. Moreover, our method sets the new state-of-the-art performance and achieves 1st place on the SemanticKITTI 4D Panoptic Segmentation leaderboard.
Knowledge distillation aims to learn a lightweight student network from a pre-trained teacher network. In practice, existing knowledge distillation methods are usually infeasible when the original training data is unavailable due to some privacy issues and data management considerations. Therefore, data-free knowledge distillation approaches proposed to collect training instances from the Internet. However, most of them have ignored the common distribution shift between the instances from original training data and webly collected data, affecting the reliability of the trained student network. To solve this problem, we propose a novel method dubbed "Knowledge Distillation between Different Distributions" (KD^ 3 ), which consists of three components. Specifically, we first dynamically select useful training instances from the webly collected data according to the combined predictions of teacher network and student network. Subsequently, we align both the weighted features and classifier parameters of the two networks for knowledge memorization. Meanwhile, we also build a new contrastive learning block called MixDistribution to generate perturbed data with a new distribution for instance alignment, so that the student network can further learn a distribution-invariant representation. Intensive experiments on various benchmark datasets demonstrate that our proposed KD^ 3 can outperform the state-of-the-art data-free knowledge distillation approaches.
We question the current evaluation practice on diffusion-based purification methods. Diffusion-based purification methods aim to remove adversarial effects from an input data point at test time. The approach gains increasing attention as an alternative to adversarial training due to the disentangling between training and testing. Well-known white-box attacks are often employed to measure the robustness of the purification. However, it is unknown whether these attacks are the most effective for the diffusion-based purification since the attacks are often tailored for adversarial training. We analyze the current practices and provide a new guideline for measuring the robustness of purification methods against adversarial attacks. Based on our analysis, we further propose a new purification strategy improving robustness compared to the current diffusion-based purification methods.
Self-supervised monocular depth estimation (SSMDE) aims at predicting the dense depth maps of monocular images, by learning to minimize a photometric loss using spatially neighboring image pairs during training. While SSMDE offers a significant scalability advantage over supervised approaches, it performs poorly on reflective surfaces as the photometric constancy assumption of the photometric loss is violated. We note that the appearance of reflective surfaces is view-dependent and often there are views of such surfaces in the training data that are not contaminated by strong specular reflections. Thus, reflective surfaces can be accurately reconstructed by aggregating the predicted depth of these views. Motivated by this observation, we propose 3D distillation: a novel training framework that utilizes the projected depth of reconstructed reflective surfaces to generate reasonably accurate depth pseudo-labels. To identify those surfaces automatically, we employ an uncertainty-guided depth fusion method, combining the smoother and more accurate projected depth on reflective surfaces and the detailed predicted depth elsewhere. In our experiments using the ScanNet and 7-Scenes datasets, we show that 3D distillation not only significantly improves the prediction accuracy, especially on the problematic surfaces, but also that it generalizes well over various underlying network architectures and to new datasets.
We introduce Zero-1-to-3, a framework for changing the camera viewpoint of an object given just a single RGB image. To perform novel view synthesis in this underconstrained setting, we capitalize on the geometric priors that large-scale diffusion models learn about natural images. Our conditional diffusion model uses a synthetic dataset to learn controls of the relative camera viewpoint, which allow new images to be generated of the same object under a specified camera transformation. Even though it is trained on a synthetic dataset, our model retains a strong zero-shot generalization ability to out-of-distribution datasets as well as in-the-wild images, including impressionist paintings. Our viewpoint-conditioned diffusion approach can further be used for the task of 3D reconstruction from a single image. Qualitative and quantitative experiments show that our method significantly outperforms stateof- the-art single-view 3D reconstruction and novel view synthesis models by leveraging Internet-scale pre-training.
Continual Segment: Towards a Single, Unified and Non-forgetting Continual Segmentation Model of 143 Whole-body Organs in CT Scans
PDF ↗Deep learning empowers the mainstream medical image segmentation methods. Nevertheless, current deep segmentation approaches are not capable of efficiently and effectively adapting and updating the trained models when new segmentation classes are incrementally added. In the real clinical environment, it can be preferred that segmentation models could be dynamically extended to segment new organs/tumors without the (re-)access to previous training datasets due to obstacles of patient privacy and data storage. This process can be viewed as a continual semantic segmentation (CSS) problem, being understudied for multi-organ segmentation. In this work, we propose a new architectural CSS learning framework to learn a single deep segmentation model for segmenting a total of 143 whole-body organs. Using the encoder/decoder network structure, we demonstrate that a continually trained then frozen encoder coupled with incrementally-added decoders can extract sufficiently representative image features for new classes to be subsequently and validly segmented, while avoiding the catastrophic forgetting in CSS. To maintain a single network model complexity, each decoder is progressively pruned using neural architecture search and teacher-student based knowledge distillation. Finally, we propose a body-part and anomaly-aware output merging module to combine organ predictions originating from different decoders and incorporate both healthy and pathological organs appearing in different datasets. Trained and validated on 3D CT scans of 2500+ patients from four datasets, our single network can segment a total of 143 whole-body organs with very high accuracy, closely reaching the upper bound performance level by training four separate segmentation models (i.e., one model per dataset/task).
Vision-Language (V-L) models trained with contrastive learning to align the visual and language modalities have been shown to be strong few-shot learners. Soft prompt learning is the method of choice for few-shot downstream adaption aiming to bridge the modality gap caused by the distribution shift induced by the new domain. While parameter-efficient, prompt learning still requires access to the model weights and can be computationally infeasible for large models with billions of parameters. To address these shortcomings, in this work, we describe a black-box method for V-L few-shot adaptation that (a) operates on pre-computed image and text features and hence works without access to the model's weights, (b) it is orders of magnitude faster at training time, (c) it is amenable to both supervised and unsupervised training, and (d) it can be even used to align image and text features computed from uni-modal models. To achieve this, we propose Linear Feature Alignment (LFA), a simple linear approach for V-L re-alignment in the target domain. LFA is initialized from a closed-form solution to a least-squares problem and then it is iteratively updated by minimizing a re-ranking loss. Despite its simplicity, our approach can even surpass soft-prompt learning methods as shown by extensive experiments on 11 image and 2 video datasets.
Reliable forecasting of the future behavior of road agents is a critical component to safe planning in autonomous vehicles. Here, we represent continuous trajectories as sequences of discrete motion tokens and cast multi-agent motion prediction as a language modeling task over this domain. Our model, MotionLM, provides several advantages: First, it does not require anchors or explicit latent variable optimization to learn multimodal distributions. Instead, we leverage a single standard language modeling objective, maximizing the average log probability over sequence tokens. Second, our approach bypasses post-hoc interaction heuristics where individual agent trajectory generation is conducted prior to interactive scoring. Instead, MotionLM produces joint distributions over interactive agent futures in a single autoregressive decoding process. In addition, the model's sequential factorization enables temporally causal conditional rollouts. The proposed approach establishes new state-of-the-art performance for multi-agent motion prediction on the Waymo Open Motion Dataset, ranking 1st on the interactive challenge leaderboard.
There has been significant progress in Masked Image Modeling (MIM). Existing MIM methods can be broadly categorized into two groups based on the reconstruction target: pixel-based and tokenizer-based approaches. The former offers a simpler pipeline and lower computational cost, but it is known to be biased toward high-frequency details. In this paper, we provide a set of empirical studies to confirm this limitation of pixel-based MIM and propose a new method that explicitly utilizes low-level features from shallow layers to aid pixel reconstruction. By incorporating this design into our base method, MAE, we reduce the wasted modeling capability of pixel-based MIM, improving its convergence and achieving non-trivial improvements across various downstream tasks. To the best of our knowledge, we are the first to systematically investigate multi-level feature fusion for isotropic architectures like the standard Vision Transformer (ViT). Notably, when applied to a smaller model (e.g., ViT-S), our method yields significant performance gains, such as 1.2% on fine-tuning, 2.8% on linear probing, and 2.6% on semantic segmentation.
We present Skill Transformer, an approach for solving long-horizon robotic tasks by combining conditional sequence modeling and skill modularity. Conditioned on egocentric and proprioceptive observations of a robot, Skill Transformer is trained end-to-end to predict both a high-level skill (e.g., navigation, picking, placing), and a whole-body low-level action (e.g., base and arm motion), using a transformer architecture and demonstration trajectories that solve the full task. It retains the composability and modularity of the overall task through a skill predictor module while reasoning about low-level actions and avoiding hand-off errors, common in modular approaches. We test Skill Transformer on an embodied rearrangement benchmark and find it performs robust task planning and low-level control in new scenarios, achieving a 2.5x higher success rate than baselines in hard rearrangement problems.
Towards Universal Image Embeddings: A Large-Scale Dataset and Challenge for Generic Image Representations
PDF ↗Fine-grained and instance-level recognition methods are commonly trained and evaluated on specific domains, in a model per domain scenario. Such an approach, however, is impractical in real large-scale applications. In this work, we address the problem of universal image embedding, where a single universal model is trained and used in multiple domains. First, we leverage existing domain-specific datasets to carefully construct a new large-scale public benchmark for the evaluation of universal image embeddings, with 241k query images, 1.4M index images and 2.8M training images across 8 different domains and 349k classes. We define suitable metrics, training and evaluation protocols to foster future research in this area. Second, we provide a comprehensive experimental evaluation on the new dataset, demonstrating that existing approaches and simplistic extensions lead to worse performance than an assembly of models trained for each domain separately. Finally, we conducted a public research competition on this topic, leveraging industrial datasets, which attracted the participation of more than 1k teams worldwide. This exercise generated many interesting research ideas and findings which we present in detail. Project webpage: https://cmp.felk.cvut.cz/univ_emb/
Closed-set 3D perception models trained on only a pre-defined set of object categories can be inadequate for safety critical applications such as autonomous driving where new object types can be encountered after deployment. In this paper, we present a multi-modal auto labeling pipeline capable of generating amodal 3D bounding boxes and tracklets for training models on open-set categories without 3D human labels. Our pipeline exploits motion cues inherent in point cloud sequences in combination with the freely available 2D image-text pairs to identify and track all traffic participants. Compared to the recent studies in this domain, which can only provide class-agnostic auto labels limited to moving objects, our method can handle both static and moving objects in the unsupervised manner and is able to output open-vocabulary semantic labels thanks to the proposed vision-language knowledge distillation. Experiments on the Waymo Open Dataset show that our approach outperforms the prior work by significant margins on various unsupervised 3D perception tasks.
Humans can easily understand a single image as depicting multiple potential objects permitting interaction. We use this skill to plan our interactions with the world and accelerate understanding new objects without engaging in interaction. In this paper, we would like to endow machines with the similar ability, so that intelligent agents can better explore the 3D scene or manipulate objects. Our approach is a transformer-based model that predicts the 3D location, physical properties and affordance of objects. To power this model, we collect a dataset with Internet videos, egocentric videos and indoor images to train and validate our approach. Our model yields strong performance on our data, and generalizes well to robotics data.
We present a new test-time optimization method for estimating dense and long-range motion from a video sequence. Prior optical flow or particle video tracking algorithms typically operate within limited temporal windows, struggling to track through occlusions and maintain global consistency of estimated motion trajectories. We propose a complete and globally consistent motion representation, dubbed OmniMotion, that allows for accurate, full-length motion estimation of every pixel in a video. OmniMotion represents a video using a quasi-3D canonical volume and performs pixel-wise tracking via bijections between local and canonical space. This representation allows us to ensure global consistency, track through occlusions, and model any combination of camera and object motion. Extensive evaluations on the TAP-Vid benchmark and real-world footage show that our approach outperforms prior state-of-the-art methods by a large margin both quantitatively and qualitatively.
We introduce a self-supervised training strategy for burst super-resolution that only uses noisy low-resolution bursts during training. Our approach eliminates the need to carefully tune synthetic data simulation pipelines, which often do not match real-world image statistics. Compared to weakly-paired training strategies, which require noisy smartphone burst photos of static scenes, paired with a clean reference obtained from a tripod-mounted DSLR camera, our approach is more scalable, and avoids the color mismatch between the smartphone and DSLR. To achieve this, we propose a new self-supervised objective that uses a forward imaging model to recover a high-resolution image from aliased high frequencies in the burst. Our approach does not require any manual tuning of the forward model's parameters; we learn them from data. Furthermore, we show our training strategy is robust to dynamic scene motion in the burst, which enables training burst super-resolution models using in-the-wild data. Extensive experiments on real and synthetic data show that, despite only using noisy bursts during training, models trained with our self-supervised strategy match, and sometimes surpass, the quality of fully-supervised baselines trained with synthetic data or weakly-paired ground-truth. Finally, we show our training strategy is general using four different burst super-resolution architectures.
Human pose transfer aims to generate a new image of the source person in a target pose. Among the existing methods, attention and flow are two of the most popular and effective approaches. Attention excels in preserving the semantic structure of the source image, which is more reflected in the low-frequency domain. Contrastively, flow is better at retaining fine-grained texture details in the high-frequency domain. To leverage the advantages of both attention and flow simultaneously, we propose Wavelet-aware Image-based Pose Transfer (WaveIPT) to fuse the attention and flow in the wavelet domain. To improve the fusion effect and avoid interference from irrelevant information between different frequencies, WaveIPT first applies Intra-scale Local Correlation (ILC) to adaptively fuse attention and flow in the same scale according to their strengths in low and high-frequency domains, and then uses Inter-scale Feature Interaction (IFI) module to explore inter-scale frequency features for effective information transfer across different scales. We further introduce an effective Progressive Flow Regularization to alleviate the challenges of flow estimation under large pose differences. Our experiments on the DeepFashion dataset demonstrate that WaveIPT achieves a new state-of-the-art in terms of FID and LPIPS, with improvements of 4.97% and 3.89%, respectively.
Recent learning-based video inpainting approaches have achieved considerable progress. However, they still cannot fully utilize semantic information within the video frames and predict improper scene layout, failing to restore clear object boundaries for mixed scenes. To mitigate this problem, we introduce a new transformer-based video inpainting technique that can exploit semantic information within the input and considerably improve reconstruction quality. In this study, we use the mixture-of-experts scheme and train multiple experts to handle mixed scenes, including various semantics. We leverage these multiple experts and produce locally (token-wise) different network parameters to achieve semantic-aware inpainting results. Extensive experiments on YouTube-VOS and DAVIS benchmark datasets demonstrate that, compared with existing conventional video inpainting approaches, the proposed method has superior performance in synthesizing visually pleasing videos with much clearer semantic structures and textures.
The main challenge of monocular 3D object detection is the accurate localization of 3D center. Motivated by a new and strong observation that this challenge can be remedied by a 3D-space local-grid search scheme in an ideal case, we propose a stage-wise approach, which combines the information flow from 2D-to-3D (3D bounding box proposal generation with a single 2D image) and 3D-to-2D (proposal verification by denoising with 3D-to-2D contexts) in a top-down manner. Specifically, we first obtain initial proposals from off-the-shelf backbone monocular 3D detectors. Then, we generate a 3D anchor space by local-grid sampling from the initial proposals. Finally, we perform 3D bounding box denoising at the 3D-to-2D proposal verification stage. To effectively learn discriminative features for denoising highly overlapped proposals, this paper presents a method of using the Perceiver I/O model to fuse the 3D-to-2D geometric information and the 2D appearance information. With the encoded latent representation of a proposal, the verification head is implemented with a self-attention module. Our method, named as MonoXiver, is generic and can be easily adapted to any backbone monocular 3D detectors. Experimental results on the well-established KITTI dataset and the challenging large-scale Waymo dataset show that MonoXiver consistently achieves improvement with limited computation overhead.