Partially Relevant Video Retrieval~(PRVR) aims to retrieve a video where a specific segment is relevant to a given text query. Typical training processes of PRVR assume a one-to-one relationship where each text query is relevant to only one video. However, we point out the inherent ambiguity between text and video content based on their conceptual scope and propose a framework that incorporates this ambiguity into the model learning process. Specifically, we propose Ambiguity-Restrained representation Learning~(ARL) to address ambiguous text-video pairs. Initially, ARL detects ambiguous pairs based on two criteria: uncertainty and similarity. Uncertainty represents whether instances include commonly shared context across the dataset, while similarity indicates pair-wise semantic overlap. Then, with the detected ambiguous pairs, our ARL hierarchically learns the semantic relationship via multi-positive contrastive learning and dual triplet margin loss. Additionally, we delve into fine-grained relationships within the video instances. Unlike typical training at the text-video level, where pairwise information is provided, we address the inherent ambiguity within frames of the same untrimmed video, which often contains multiple contexts. This allows us to further enhance learning at the text-frame level. Lastly, we propose cross-model ambiguity detection to mitigate the error propagation that occurs when a single model is employed to detect ambiguous pairs for its training. With all components combined, our proposed method demonstrates its effectiveness in PRVR.

Partially Relevant Video Retrieval (PRVR) addresses the challenges of text-to-video retrieval in real-world scenarios where untrimmed videos are prevalent. Traditional PRVR methods encode videos at two feature scales: (1) frame-level to capture fine details, and (2) clip-level to recognize broader content. However, these approaches align both scales with a single sentence representation, leading to suboptimal performance. In particular, we point out the level mismatch in aligning frame-level video features with a sentence representation, as the entire meaning of a sentence contains broader and more diverse content than what frame-level features can encode. This misalignment causes frame-level features to capture broader contexts and overlook local fine details. To tackle this issue, we propose a framework that represents a sentence as a set of multiple components, where each component aligns with frame-level semantics. Specifically, we introduce Semantic-Decomposed Matching (SDM) to adjust the granularity of the text description to match them with frame-level video features. In addition to the matching process, we develop the Adaptive Local Aggregator (ALA) to enhance video encoding in capturing finer local details, ensuring precise text-video alignment at the frame level. ALA adaptively integrates multi-scale local details within short temporal spans obtained by enforcing a strict temporal aggregation range. Finally, we reinforce detailed encoding at the frame level with newly designed objectives for both modalities. Extensive experiments integrating our framework with existing clip branches demonstrate its effectiveness and applicability, highlighting significant improvements in PRVR performance.

Temporal Action Detection (TAD) is fundamental yet challenging for real-world video applications. Leveraging the unique benefits of transformers, various DETR-based approaches have been adopted in TAD. However, it has recently been identified that the attention collapse in self-attention causes the performance degradation of DETR for TAD. Building upon previous research, this paper newly addresses the attention collapse problem in cross-attention within DETR-based TAD methods. Moreover, our findings reveal that cross-attention exhibits patterns distinct from predictions, indicating a short-cut phenomenon. To resolve this, we propose a new framework, Prediction-Feedback DETR (Pred-DETR), which utilizes predictions to restore the collapse and align the cross- and self-attention with predictions. Specifically, we devise novel prediction-feedback objectives using guidance from the relations of the predictions. As a result, Pred-DETR significantly alleviates the collapse and achieves state-of-the-art performance among DETR-based methods on various challenging benchmarks including THUMOS14, ActivityNet-v1.3, HACS, and FineAction.

We propose Foreground-Covering Prototype Generation and Matching to resolve Few-Shot Segmentation (FSS), which aims to segment target regions in unlabeled query images based on labeled support images. Unlike previous research, which typically estimates target regions in the query using support prototypes and query pixels, we utilize the relationship between support and query prototypes. To achieve this, we utilize two complementary features: SAM Image Encoder features for pixel aggregation and ResNet features for class consistency. Specifically, we construct support and query prototypes with SAM features and distinguish query prototypes of target regions based on ResNet features. For the query prototype construction, we begin by roughly guiding foreground regions within SAM features using the conventional pseudo-mask, then employ iterative cross-attention to aggregate foreground features into learnable tokens. Here, we discover that the cross-attention weights can effectively alternate the conventional pseudo-mask. Therefore, we use the attention-based pseudo-mask to guide ResNet features to focus on the foreground, then infuse the guided ResNet feature into the learnable tokens to generate class-consistent query prototypes. The generation of the support prototype is conducted symmetrically to that of the query one, with the pseudo-mask replaced by the ground-truth mask. Finally, we compare these query prototypes with support ones to generate prompts, which subsequently produce object masks through the SAM Mask Decoder. Our state-of-the-art performances on various datasets validate the effectiveness of the proposed method for FSS.

This work tackles the fidelity objective in the perceptual super-resolution (SR) task. Specifically, we address the shortcomings of pixel-level ℒploss (ℒpix) in the GAN-based SR framework. Since ℒpixis known to have a trade-off relationship against perceptual quality, prior methods often multiply a small scale factor or utilize low-pass filters. However, this work shows that these circumventions fail to address the fundamental factor that induces blurring. Accordingly, we focus on two points: 1) precisely discriminating the subcomponent of ℒpixthat contributes to blurring, and 2) only guiding based on the factor that is free from this trade-off relationship. We show that they can be achieved in a surprisingly simple manner, with an Auto-Encoder (AE) pretrained with ℒpix. Accordingly, we propose the Auto-Encoded Supervision for Optimal Penalization loss (ℒAESOP), a novel loss function that measures distance in the AE space1, instead of the raw pixel space. By simply substituting ℒpixwith ℒAESOP, we can provide effective reconstruction guidance without compromising perceptual quality. Designed for simplicity, our method enables easy integration into existing SR frameworks. Extensive experiments demonstrate the effectiveness of AESOP.

Few-Shot Action Recognition (FSAR) aims to train a model with only a few labeled video instances. A key challenge in FSAR is handling divergent narrative trajectories for precise video matching. While the frame- and tuple-level alignment approaches have been promising, their methods heavily rely on pre-defined and length-dependent alignment units (e.g., frames or tuples), which limits flexibility for actions of varying lengths and speeds. In this work, we introduce a novel TEmporal Alignment-Free Matching (TEAM) approach, which eliminates the need for temporal units in action representation and brute-force alignment during matching. Specifically, TEAM represents each video with a fixed set of pattern tokens that capture globally discriminative clues within the video instance regardless of action length or speed, ensuring its flexibility. Furthermore, TEAM is inherently efficient, using token-wise comparisons to measure similarity between videos, unlike existing methods that rely on pairwise comparisons for temporal alignment. Additionally, we propose an adaptation process that identifies and removes common information across classes, establishing clear boundaries even between novel categories. Extensive experiments demonstrate the effectiveness of TEAM. Codes are available at github.com/leesb7426/TEAM.

Recent advances in text-to-image generative models have enabled numerous practical applications, including subject-driven generation, which fine-tunes pretrained models to capture subject semantics from only a few examples. While diffusion-based models produce high-quality images, their extensive denoising steps result in significant computational overhead, limiting real-world applicability. Visual autoregressive (VAR) models, which predict next-scale tokens rather than spatially adjacent ones, offer significantly faster inference suitable for practical deployment. In this paper, we propose the first VAR-based approach for subject-driven generation. However, naive fine-tuning VAR leads to computational overhead, language drift, and reduced diversity. To address these challenges, we introduce selective layer tuning to reduce complexity and prior distillation to mitigate language drift. Additionally, we found that the early stages have a greater influence on the generation of subject than the latter stages, which merely synthesize minor details. Based on this finding, we propose scale-wise weighted tuning, which prioritizes coarser resolutions for promoting the model to focus on the subject-relevant information instead of local details. Extensive experiments validate that our method significantly outperforms diffusion-based baselines across various metrics and demonstrates its practical usage.

In a retrieval system, simultaneously achieving search accuracy and efficiency is inherently challenging. This challenge is particularly pronounced in partially relevant video retrieval (PRVR), where incorporating more diverse context representations at varying temporal scales for each video enhances accuracy but increases computational and memory costs. To address this dichotomy, we propose a prototypical PRVR framework that encodes diverse contexts within a video into a fixed number of prototypes. We then introduce several strategies to enhance text association and video understanding within the prototypes, along with an orthogonal objective to ensure that the prototypes capture a diverse range of content. To keep the prototypes searchable via text queries while accurately encoding video contexts, we implement cross- and uni-modal reconstruction tasks. The cross-modal reconstruction task aligns the prototypes with textual features within a shared space, while the uni-modal reconstruction task preserves all video contexts during encoding. Additionally, we employ a video mixing technique to provide weak guidance to further align prototypes and associated textual representations. Extensive evaluations on TVR, ActivityNet-Captions, and QVHighlights validate the effectiveness of our approach without sacrificing efficiency.

Facilitating an entity's interaction with objects requires accurately identifying parts that afford specific actions. Weakly supervised affordance grounding (WSAG) seeks to imitate human learning from third-person demonstrations, where humans intuitively grasp functional parts without needing pixel-level annotations. To achieve this, grounding is typically learned using a shared classifier across images from different perspectives, along with distillation strategies incorporating part discovery process. However, since affordance-relevant parts are not always easily distinguishable, models primarily rely on classification, often focusing on common class-specific patterns that are unrelated to affordance. To address this limitation, we move beyond isolated part-level learning by introducing selective prototypical and pixel contrastive objectives that adaptively learn affordance-relevant cues at both the part and object levels, depending on the granularity of the available information. Initially, we find the action-associated objects in both egocentric (object-focused) and exocentric (third-person example) images by leveraging CLIP. Then, by cross-referencing the discovered objects of complementary views, we excavate the precise part-level affordance clues in each perspective. By consistently learning to distinguish affordance-relevant regions from affordance-irrelevant background context, our approach effectively shifts activation from irrelevant areas toward meaningful affordance cues. Experimental results demonstrate the effectiveness of our method. Codes are available at github.com/hynnsk/SelectiveCL.

Recent advances in text-based image editing have motivated the extension of these techniques into the 3D domain. However, existing methods typically apply 2D diffusion models independently to multiple viewpoints, resulting in significant artifacts, most notably the Janus problem, due to inconsistencies across edited views. To address this, we propose a novel approach termed DFFSplat, which integrates a 3D-consistent diffusion feature field into the editing pipeline. By rendering and injecting these 3D-consistent structural features into intermediate layers of a 2D diffusion model, our method effectively enforces geometric alignment and semantic coherence across views. However, averaging 3D features during the feature field learning process can lead to the loss of fine texture details. To overcome this, we introduce a dual-encoder architecture to disentangle view-independent structural information from view-dependent appearance details. By encoding only the disentangled structure into the 3D field and injecting it during 2D editing, our method produces semantically and multi-view coherent edited images while maintaining high text fidelity. Additionally, we employ a time-invariance objective to ensure consistency across diffusion timesteps, enhancing the stability of learned representations. Experimental results demonstrate that our method achieves state-of-the-art performance in terms of text-fidelity, and better preserves structural and semantic consistency compared to existing approaches.

Partially Relevant Video Retrieval (PRVR) seeks videos where only part of the content matches a text query. Existing methods treat every annotated text–video pair as a positive and all others as negatives, ignoring the rich semantic variation both within a single video and across different videos. Consequently, embeddings of both queries and their corresponding video‐clip segments for distinct events within the same video collapse together, while embeddings of semantically similar queries and segments from different videos are driven apart. This limits retrieval performance when videos contain multiple, diverse events. This paper addresses the aforementioned problems, termed as semantic collapse, in both the text and video embedding spaces. We first introduce Text Correlation Preservation Learning, which preserves the semantic relationships encoded by the foundation model across text queries. To address collapse in video embeddings, we propose Cross-Branch Video Alignment (CBVA), a contrastive alignment method that disentangles hierarchical video representations across temporal scales. Subsequently, we introduce order-preserving token merging and adaptive CBVA to enhance alignment by producing video segments that are internally coherent yet mutually distinctive. Extensive experiments on PRVR benchmarks demonstrate that our framework effectively prevents semantic collapse and substantially improves retrieval accuracy.

The difficulty of fine-grained image classification mainly comes from a shared overall appearance across classes. Thus, recognizing discriminative details, such as the eyes and beaks of birds, is a key to the task. However, this is particularly challenging when training data is limited. To address this, we propose Task Discrepancy Maximization (TDM), a task-oriented channel attention method tailored for fine-grained few-shot classification with two novel modules Support Attention Module (SAM) and Query Attention Module (QAM). SAM highlights channels encoding class-wise discriminative features, while QAM assigns higher weights to object-relevant channels of the query. Based on these submodules, TDM produces task-adaptive features by focusing on channels encoding class-discriminative details and possessed by the query at the same time, for accurate class-sensitive similarity measure between support and query instances. While TDM influences high-level feature maps by task-adaptive calibration of channel-wise importance, we further introduce Instance Attention Module (IAM) operating in intermediate layers of feature extractors to instance-wisely highlight object-relevant channels, by extending QAM. The merits of TDM and IAM and their complementary benefits are experimentally validated in fine-grained few-shot classification tasks. Moreover, IAM is also effective in coarse-grained and cross-domain few-shot classifications.

Zero-Shot Object Counting~(ZSOC) aims to count referred instances of arbitrary classes in a query image without human-annotated exemplars. To deal with ZSOC, preceding studies proposed a two-stage pipeline: discovering exemplars and counting. However, there remains a challenge of vulnerability to error propagation of the sequentially designed two-stage process. In this work, we propose an one-stage baseline, Visual-Language Baseline (VLBase), exploring the implicit association of the semantic-patch embeddings of CLIP. Subsequently, we extend the VLBase to Visual-language Counter (VLCounter) by incorporating three modules devised to tailor VLBase for object counting. First, we introduce Semantic-conditioned Prompt Tuning (SPT) within the image encoder to acquire target-highlighted representations. Second, Learnable Affine Transformation (LAT) is employed to translate the semantic-patch similarity map to be appropriate for the counting task. Lastly, we transfer the layer-wisely encoded features to the decoder through Segment-aware Skip Connection (SaSC) to keep the generalization capability for unseen classes. Through extensive experiments on FSC147, CARPK, and PUCPR+, we demonstrate the benefits of our end-to-end framework, VLCounter. Code is available at https://github.com/seunggu0305/VLCounter

Recent deep-learning-based single image super-resolution (SISR) methods have shown impressive performance whereas typical methods train their networks by minimizing the pixel-wise distance with respect to a given high-resolution (HR) image. However, despite the basic training scheme being the predominant choice, its use in the context of ill-posed inverse problems has not been thoroughly investigated. In this work, we aim to provide a better comprehension of the underlying constituent by decomposing target HR images into two subcomponents: (1) the optimal centroid which is the expectation over multiple potential HR images, and (2) the inherent noise defined as the residual between the HR image and the centroid. Our findings show that the current training scheme cannot capture the ill-posed nature of SISR and becomes vulnerable to the inherent noise term, especially during early training steps. To tackle this issue, we propose a novel optimization method that can effectively remove the inherent noise term in the early steps of vanilla training by estimating the optimal centroid and directly optimizing toward the estimation. Experimental results show that the proposed method can effectively enhance the stability of vanilla training, leading to overall performance gain. Codes are available at github.com/2minkyulee/ECO.

Few-shot segmentation aims to accurately segment novel target objects within query images using only a limited number of annotated support images. The recent works exploit support background as well as its foreground to precisely compute the dense correlations between query and support. However, they overlook the characteristics of the background that generally contains various types of objects. In this paper, we highlight this characteristic of background which can bring problematic cases as follows: (1) when the query and support backgrounds are dissimilar and (2) when objects in the support background are similar to the target object in the query. Without any consideration of the above cases, adopting the entire support background leads to a misprediction of the query foreground as background. To address this issue, we propose Task-disruptive Background Suppression(TBS), a module to suppress those disruptive support background features based on two spatial-wise scores: query-relevant and target-relevant scores. The former aims to mitigate the impact of unshared features solely existing in the support background, while the latter aims to reduce the influence of target-similar support background features. Based on these two scores, we define a query background relevant score that captures the similarity between the backgrounds of the query and the support, and utilize it to scale support background features to adaptively restrict the impact of disruptive support backgrounds. Our proposed method achieves state-of-the-art performance on standard few-shot segmentation benchmarks. Our official code is available at github.com/SuhoPark0706/TBSNet.

In this paper, we first investigate a visual quality degradation problem observed in recent high-resolution virtual try-on approach. The tendency is empirically found that the textures of clothes are squeezed at the sleeve, as visualized in the upper row of Fig.1(a). A main reason for the issue arises from a gradient conflict between two popular losses, the Total Variation (TV) and adversarial losses. Specifically, the TV loss aims to disconnect boundaries between the sleeve and torso in a warped clothing mask, whereas the adversarial loss aims to combine between them. Such contrary objectives feedback the misaligned gradients to a cascaded appearance flow estimation, resulting in undesirable squeezing artifacts. To reduce this, we propose a Sequential Deformation (SD-VITON) that disentangles the appearance flow prediction layers into TV objective-dominant (TVOB) layers and a task-coexistence (TACO) layer. Specifically, we coarsely fit the clothes onto a human body via the TVOB layers, and then keep on refining via the TACO layer. In addition, the bottom row of Fig.1(a) shows a different type of squeezing artifacts around the waist. To address it, we further propose that we first warp the clothes into a tucked-out shirts style, and then partially erase the texture from the warped clothes without hurting the smoothness of the appearance flows. Experimental results show that our SD-VITON successfully resolves both types of artifacts and outperforms the baseline methods. Source code will be available at https://github.com/SHShim0513/SD-VITON.

Despite the impressive generative capabilities of diffusion models, existing diffusion model-based style transfer methods require inference-stage optimization (e.g. fine-tuning or textual inversion of style) which is time-consuming, or fails to leverage the generative ability of large-scale diffusion models. To address these issues, we introduce a novel artistic style transfer method based on a pre-trained large-scale diffusion model without any optimization. Specifically, we manipulate the features of self-attention layers as the way the cross-attention mechanism works; in the generation process, substituting the key and value of content with those of style image. This approach provides several desirable characteristics for style transfer including 1) preservation of content by transferring similar styles into similar image patches and 2) transfer of style based on similarity of local texture (e.g. edge) between content and style images. Furthermore, we introduce query preservation and attention temperature scaling to mitigate the issue of disruption of original content, and initial latent Adaptive Instance Normalization (AdaIN) to deal with the disharmonious color (failure to transfer the colors of style). Our experimental results demonstrate that our proposed method surpasses state-of-the-art methods in both conventional and diffusion-based style transfer baselines. Codes are available at github.com/jiwoogit/StyleID.

StyleGAN has shown remarkable performance in unconditional image generation. However, its high computational cost poses a significant challenge for practical applications. Although recent efforts have been made to compress Style-GAN while preserving its performance, existing compressed models still lag behind the original model, particularly in terms of sample diversity. To overcome this, we propose a novel channel pruning method that leverages varying sensitivities of channels to latent vectors, which is a key factor in sample diversity. Specifically, by assessing channel importance based on their sensitivities to latent vector perturbations, our method enhances the diversity of samples in the compressed model. Since our method solely focuses on the channel pruning stage, it has complementary benefits with prior training schemes without additional training cost. Ex-tensive experiments demonstrate that our method significantly enhances sample diversity across various datasets. Moreover, in terms of FID scores, our method not only sur-passes state-of-the-art by a large margin but also achieves comparable scores with only half training iterations. Codes are available at github.com/jiwoogit/DCP-GAN.

Most advances in 3D Generative Adversarial Networks (3D GANs) largely depend on ray casting-based volume rendering, which incurs demanding rendering costs. One promising alternative is rasterization-based 3D Gaussian Splatting (3D-GS), providing a much faster rendering speed and explicit 3D representation. In this paper, we exploit Gaussian as a 3D representation for 3D GANs by leveraging its efficient and explicit characteristics. However, in an adversarial framework, we observe that a na\"ive generator architecture suffers from training instability and lacks the capability to adjust the scale of Gaussians. This leads to model divergence and visual artifacts due to the absence of proper guidance for initialized positions of Gaussians and densification to manage their scales adaptively. To address these issues, we introduce GSGAN, a generator architecture with a hierarchical multi-scale Gaussian representation that effectively regularizes the position and scale of generated Gaussians. Specifically, we design a hierarchy of Gaussians where finer-level Gaussians are parameterized by their coarser-level counterparts; the position of finer-level Gaussians would be located near their coarser-level counterparts, and the scale would monotonically decrease as the level becomes finer, modeling both coarse and fine details of the 3D scene. Experimental results demonstrate that ours achieves a significantly faster rendering speed (×100) compared to state-of-the-art 3D consistent GANs with comparable 3D generation capability.

Multi-scene absolute pose regression addresses the demand for fast and memory-efficient camera pose estimation across various real-world environments. Nowadays, transformer-based model has been devised to regress the camera pose directly in multi-scenes. Despite its potential, transformer encoders are underutilized due to the collapsed self-attention map, having low representation capacity. This work highlights the problem and investigates it from a new perspective: distortion of query-key embedding space. Based on the statistical analysis, we reveal that queries and keys are mapped in completely different spaces while only a few keys are blended into the query region. This leads to the collapse of the self-attention map as all queries are considered similar to those few keys. Therefore, we propose simple but effective solutions to activate self-attention. Concretely, we present an auxiliary loss that aligns queries and keys, preventing the distortion of query-key space and encouraging the model to find global relations by self-attention. In addition, the fixed sinusoidal positional encoding is adopted instead of undertrained learnable one to reflect appropriate positional clues into the inputs of self-attention. As a result, our approach resolves the aforementioned problem effectively, thus outperforming existing methods in both outdoor and indoor scenes.

In few-shot generative model adaptation, the model for target domain is prone to the mode-collapse. Recent studies attempted to mitigate the problem by matching the relationship among samples generated from the same latent codes in source and target domains. The objective is further extended to image patch-level to transfer the spatial correlation within an instance. However, the patch-level approach assumes the consistency of spatial structure between source and target domains. For example, the positions of eyes in two domains are almost identical. Thus, it can bring visual artifacts if source and target domain images are not nicely aligned. In this paper, we propose a few-shot generative model adaptation method free from such assumption, based on a motivation that generative models are progressively adapting from the source domain to the target domain. Such progressive changes allow us to identify semantically coherent image regions between instances generated by models at a neighboring training iteration to consider the spatial correlation. We also propose an importance-based patch selection strategy to reduce the complexity of patch-level correlation matching. Our method shows the state-of-the-art few-shot domain adaptation performance in the qualitative and quantitative evaluations.

A dramatic increase in real-world video volume with extremely diverse and emerging topics naturally forms a long-tailed video distribution in terms of their categories, and it spotlights the need for Video Long-Tailed Recognition (VLTR). In this work, we summarize the challenges in VLTR and explore how to overcome them. The challenges are: (1) it is impractical to re-train the whole model for high-quality features, (2) acquiring frame-wise labels requires extensive cost, and (3) long-tailed data triggers biased training. Yet, most existing works for VLTR unavoidably utilize image-level features extracted from pretrained models which are task-irrelevant, and learn by video-level labels. Therefore, to deal with such (1) task-irrelevant features and (2) video-level labels, we introduce two complementary learnable feature aggregators. Learnable layers in each aggregator are to produce task-relevant representations, and each aggregator is to assemble the snippet-wise knowledge into a video representative. Then, we propose Minority-Oriented Vicinity Expansion (MOVE) that explicitly leverages the class frequency into approximating the vicinity distributions to alleviate (3) biased training. By combining these solutions, our approach achieves state-of-the-art results on large-scale VideoLT and synthetically induced Imbalanced-MiniKinetics200. With VideoLT features from ResNet-50, it attains 18% and 58% relative improvements on head and tail classes over the previous state-of-the-art method, respectively. Code and dataset are available at https://github.com/wjun0830/MOVE.

Images taken in low light conditions typically contain distracting noise, and eliminating such noise is a crucial computer vision problem. Additional photos captured with a camera flash can guide an image denoiser to preserve edges since the flash images often contain fine details with reduced noise. Nonetheless, a denoiser can be misled by inconsistent flash images, which have image structures (e.g., edges) that do not exist in no-flash images. Unfortunately, this disparity frequently occurs as the flash/no-flash pairs are taken in different light conditions. We propose a learning-based technique that robustly fuses the image pairs while considering their inconsistency. Our framework infers consistent flash image patches locally, which have similar image structures with the ground truth, and denoises no-flash images using the inferred ones via a combination model. We demonstrate that our technique can produce more robust results than state-of-the-art methods, given various flash/no-flash pairs with inconsistent image structures. The source code is available at https://github.com/CGLab-GIST/RIDFnF.

Recently, video moment retrieval and highlight detection (MR/HD) are being spotlighted as the demand for video understanding is drastically increased. The key objective of MR/HD is to localize the moment and estimate clip-wise accordance level, i.e., saliency score, to the given text query. Although the recent transformer-based models brought some advances, we found that these methods do not fully exploit the information of a given query. For example, the relevance between text query and video contents is sometimes neglected when predicting the moment and its saliency. To tackle this issue, we introduce Query-Dependent DETR (QD-DETR), a detection transformer tailored for MR/HD. As we observe the insignificant role of a given query in transformer architectures, our encoding module starts with cross-attention layers to explicitly inject the context of text query into video representation. Then, to enhance the model's capability of exploiting the query information, we manipulate the video-query pairs to produce irrelevant pairs. Such negative (irrelevant) video-query pairs are trained to yield low saliency scores, which in turn, encourages the model to estimate precise accordance between query-video pairs. Lastly, we present an input-adaptive saliency predictor which adaptively defines the criterion of saliency scores for the given video-query pairs. Our extensive studies verify the importance of building the query-dependent representation for MR/HD. Specifically, QD-DETR outperforms state-of-the-art methods on QVHighlights, TVSum, and Charades-STA datasets. Codes are available at github.com/wjun0830IQD-DETR.

The inverted index is a widely used data structure to avoid the infeasible exhaustive search. It accelerates retrieval significantly by splitting the database into multiple disjoint sets and restricts distance computation to a small fraction of the database. Moreover, it even improves search quality by allowing quantizers to exploit the compact distribution of residual vector space. However, we firstly point out a problem that an existing deep learning-based quantizer hardly benefits from the residual vector space, unlike conventional shallow quantizers. To cope with this problem, we introduce a novel disentangled representation learning for unsupervised neural quantization. Similar to the concept of residual vector space, the proposed method enables more compact latent space by disentangling information of the inverted index from the vectors. Experimental results on large-scale datasets confirm that our method outperforms the state-of-the-art retrieval systems by a large margin.

Dramatic demand for manpower to label pixel-level annotations triggered the advent of unsupervised semantic segmentation. Although the recent work employing the vision transformer (ViT) backbone shows exceptional performance, there is still a lack of consideration for task-specific training guidance and local semantic consistency. To tackle these issues, we leverage contrastive learning by excavating hidden positives to learn rich semantic relationships and ensure semantic consistency in local regions. Specifically, we first discover two types of global hidden positives, task-agnostic and task-specific ones for each anchor based on the feature similarities defined by a fixed pre-trained back-bone and a segmentation head-in-training, respectively. A gradual increase in the contribution of the latter induces the model to capture task-specific semantic features. In addition, we introduce a gradient propagation strategy to learn semantic consistency between adjacent patches, under the inherent premise that nearby patches are highly likely to possess the same semantics. Specifically, we add the loss propagating to local hidden positives, semantically similar nearby patches, in proportion to the predefined similarity scores. With these training schemes, our proposed method achieves new state-of-the-art (SOTA) results in COCO-stuff, Cityscapes, and Potsdam-3 datasets. Our code is available at: https://github.com/hynnsk/HP.

Temporal Action Detection (TAD) is challenging but fundamental for real-world video applications. Recently, DETR-based models have been devised for TAD but have not performed well yet. In this paper, we point out the problem in the self-attention of DETR for TAD; the attention modules focus on a few key elements, called temporal collapse problem. It degrades the capability of the encoder and decoder since their self-attention modules play no role. To solve the problem, we propose a novel framework, Self-DETR, which utilizes cross-attention maps of the decoder to reactivate self-attention modules. We recover the relationship between encoder features by simple matrix multi-plication of the cross-attention map and its transpose. Likewise, we also get the information within decoder queries. By guiding collapsed self-attention maps with the guidance map calculated, we settle down the temporal collapse of self-attention modules in the encoder and decoder. Our extensive experiments demonstrate that Self-DETR resolves the temporal collapse problem by keeping high diversity of attention over all layers. Moreover, it is validated that our simple framework achieves a new state-of-the-art performance on THUMOS14 and outperforms all the DETR-based approaches on ActivityNet-v1.3.

Videos contain motions of various speeds. For example, the motions of one's head and mouth differ in terms of speed - the head being relatively stable and the mouth moving rapidly as one speaks. Despite its diverse nature, previous video GANs generate video based on a single unified motion representation without considering the aspect of speed. In this paper, we propose a frequency-based motion representation for video GANs to realize the concept of speed in video generation process. In detail, we represent motions as continuous sinusoidal signals of various frequencies by introducing a coordinate-based motion generator. We show, in that case, frequency is highly related to the speed of motion. Based on this observation, we present frequency-aware weight modulation that enables manipulation of motions within a specific range of speed, which could not be achieved with the previous techniques. Extensive experiments validate that the proposed method outperforms state-of-the-art video GANs in terms of generation quality by its capability to model various speed of motions. Furthermore, we also show that our temporally continuous representation enables to further synthesize intermediate and future frames of generated videos.

Recognizing discriminative details such as eyes and beaks is important for distinguishing fine-grained classes since they have similar overall appearances. In this regard, we introduce Task Discrepancy Maximization (TDM), a simple module for fine-grained few-shot classification. Our objective is to localize the class-wise discriminative regions by highlighting channels encoding distinct information of the class. Specifically, TDM learns task-specific channel weights based on two novel components: Support Attention Module (SAM) and Query Attention Module (QAM). SAM produces a support weight to represent channel-wise discriminative power for each class. Still, since the SAM is basically only based on the labeled support sets, it can be vulnerable to bias toward such support set. Therefore, we propose QAM which complements SAM by yielding a query weight that grants more weight to object-relevant channels for a given query image. By combining these two weights, a class-wise task-specific channel weight is defined. The weights are then applied to produce task-adaptive feature maps more focusing on the discriminative details. Our experiments validate the effectiveness of TDM and its complementary benefits with prior methods in fine- grained few-shot classification.

In this paper, we first investigate the class-wise visual quality imbalance problem of scene images generated by GANs. The tendency is empirically found that the class-wise visual qualities are highly correlated with the dominance of object classes in the training data in terms of their scales and appearance frequencies. Specifically, the synthesized qualities of small and less frequent object classes tend to be low. To address this, we propose a novel attention module, Local Attention Pyramid (LAP) module tailored for scene image synthesis, that encourages GANs to generate diverse object classes in a high quality by explicit spread of high attention scores to local regions, since objects in scene images are scattered over the entire images. Moreover, our LAP assigns attention scores in a multiple scale to reflect the scale diversity of various objects. The experimental evaluations on three different datasets show consistent improvements in Frechet Inception Distance (FID) and Frechet Segmentation Distance (FSD) over the state-of-the-art baselines. Furthermore, we apply our LAP module to various GANs methods to demonstrate a wide applicability of our LAP module.

A video can be represented by the composition of appearance and motion. Appearance (or content) expresses the information invariant throughout time, and motion describes the time-variant movement. Here, we propose self-supervised approaches for video Generative Adversarial Networks (GANs) to achieve the appearance consistency and motion coherency in videos. Specifically, the dual discriminators for image and video individually learn to solve their own pretext tasks; appearance contrastive learning and temporal structure puzzle. The proposed tasks enable the discriminators to learn representations of appearance and temporal context, and force the generator to synthesize videos with consistent appearance and natural flow of motions. Extensive experiments in facial expression and human action public benchmarks show that our method outperforms the state-of-the-art video GANs. Moreover, consistent improvements regardless of the architecture of video GANs confirm that our framework is generic.

Typical domain adaptation techniques aim to transfer the knowledge learned from a label-rich source domain to a label-scarce target domain in the same label space. However, it is often hard to get even the unlabeled target domain data of a task of interest. In such a case, we can capture the domain shift between the source domain and target domain from an unseen task and transfer it to the task of interest, which is known as zero-shot domain adaptation (ZSDA). Most of existing state-of-the-art methods for ZSDA attempted to generate target domain data. However, training such generative models causes significant computational overhead and is hardly optimized. In this paper, we propose a novel ZSDA method that learns a task-agnostic domain shift by collaborative training of domain-invariant semantic features and task-invariant domain features via adversarial learning. Meanwhile, the spatial attention map is learned from disentangled feature representations to selectively emphasize the domain-specific salient parts of the domain-invariant features. Experimental results show that our ZSDA method achieves state-of-the-art performance on several benchmarks.

The inverted index is one of the most commonly used structures for non-exhaustive nearest neighbor search on large-scale datasets. It allows a significant factor of acceleration by a reduced number of distance computations with only a small fraction of the database. In particular, the inverted index enables the product quantization (PQ) to learn their codewords in the residual vector space. The quantization error of the PQ can be substantially improved in such combination since the residual vector space is much more quantization-friendly thanks to their compact distribution compared to the original data. In this paper, we first raise an unremarked but crucial question; why the inverted index and the product quantizer are optimized separately even though they are closely related? For instance, changes on the inverted index distort the whole residual vector space. To address the raised question, we suggest a joint optimization of the coarse and fine quantizers by substituting the original objective of the coarse quantizer to end-to-end quantization distortion. Moreover, our method is generic and applicable to different combinations of coarse and fine quantizers such as inverted multi-index and optimized PQ.

Approximate K-nearest neighbor search is a fundamental problem in computer science. The problem is especially important for high-dimensional and large-scale data. Recently, many techniques encoding high-dimensional data to compact codes have been proposed. The product quantization and its variations that encode the cluster index in each subspace have been shown to provide impressive accuracy. In this paper, we explore a simple question: is it best to use all the bit-budget for encoding a cluster index? We have found that as data points are located farther away from the cluster centers, the error of estimated distance becomes larger. To address this issue, we propose a novel compact code representation that encodes both the cluster index and quantized distance between a point and its cluster center in each subspace by distributing the bit-budget. We also propose two distance estimators tailored to our representation. We further extend our method to encode global residual distances in the original space. We have evaluated our proposed methods on benchmarks consisting of GIST, VLAD, and CNN features. Our extensive experiments show that the proposed methods significantly and consistently improve the search accuracy over other tested techniques. This result is achieved mainly because our methods accurately estimate distances.

In this paper, we introduce a novel shortlist computation algorithm for approximate, high-dimensional nearest neighbor search. Our method relies on a novel distance estimator: the residual-aware distance estimator, that accounts for the residual distances of data points to their respective quantized centroids, and uses it for accurate short-list computation. Furthermore, we perform the residual-aware distance estimation with little additional memory and computational cost through simple pre-computation methods for inverted index and multi-index schemes. Because it modifies the initial shortlist collection phase, our new algorithm is applicable to most inverted indexing methods that use vector quantization. We have tested the proposed method with the inverted index and multi-index on a diverse set of benchmarks including up to one billion data points with varying dimensions, and found that our method robustly improves the accuracy of shortlists (up to 127% relatively higher) over the state-of-the-art techniques with a comparable or even faster computational cost.

Many binary code embedding schemes have been actively studied recently, since they can provide efficient similarity search, and compact data representations suitable for handling large scale image databases. Existing binary code embedding techniques encode high-dimensional data by using hyperplane-based hashing functions. In this paper we propose a novel hypersphere-based hashing function, spherical hashing, to map more spatially coherent data points into a binary code compared to hyperplane-based hashing functions. We also propose a new binary code distance function, spherical Hamming distance, tailored for our hypersphere-based binary coding scheme, and design an efficient iterative optimization process to achieve both balanced partitioning for each hash function and independence between hashing functions. Furthermore, we generalize spherical hashing to support various similarity measures defined by kernel functions. Our extensive experiments show that our spherical hashing technique significantly outperforms state-of-the-art techniques based on hyperplanes across various benchmarks with sizes ranging from one to 75 million of GIST, BoW and VLAD descriptors. The performance gains are consistent and large, up to 100 percent improvements over the second best method among tested methods. These results confirm the unique merits of using hyperspheres to encode proximity regions in high-dimensional spaces. Finally, our method is intuitive and easy to implement.

Many binary code embedding techniques have been proposed for large-scale approximate nearest neighbor search in computer vision. Recently, product quantization that encodes the cluster index in each subspace has been shown to provide impressive accuracy for nearest neighbor search. In this paper, we explore a simple question: is it best to use all the bit budget for encoding a cluster index in each subspace? We have found that as data points are located farther away from the centers of their clusters, the error of estimated distances among those points becomes larger. To address this issue, we propose a novel encoding scheme that distributes the available bit budget to encoding both the cluster index and the quantized distance between a point and its cluster center. We also propose two different distance metrics tailored to our encoding scheme. We have tested our method against the-state-of-the-art techniques on several well-known benchmarks, and found that our method consistently improves the accuracy over other tested methods. This result is achieved mainly because our method accurately estimates distances between two data points with the new binary codes and distance metric.

We present a fast, yet accurate k-nearest neighbor search algorithm for high-dimensional sampling-based motion planners. Our technique is built on top of Locality Sensitive Hashing (LSH), but is extended to support arbitrary distance metrics used for motion planning problems and adapt irregular distributions of samples generated in the configuration space. To enable such novel characteristics our method embeds samples generated in the configuration space into a simple l2norm space by using pivot points. We then implicitly define Voronoi regions and use local LSHs with varying quantization factors for those Voronoi regions. We have applied our method and other prior techniques to high-dimensional motion planning problems. Our method is able to show performance improvement by a factor of up to three times even with higher accuracy over prior, approximate nearest neighbor search techniques.

Many binary code encoding schemes based on hashing have been actively studied recently, since they can provide efficient similarity search, especially nearest neighbor search, and compact data representations suitable for handling large scale image databases in many computer vision problems. Existing hashing techniques encode high-dimensional data points by using hyperplane-based hashing functions. In this paper we propose a novel hypersphere-based hashing function, spherical hashing, to map more spatially coherent data points into a binary code compared to hyperplane-based hashing functions. Furthermore, we propose a new binary code distance function, spherical Hamming distance, that is tailored to our hypersphere-based binary coding scheme, and design an efficient iterative optimization process to achieve balanced partitioning of data points for each hash function and independence between hashing functions. Our extensive experiments show that our spherical hashing technique significantly outperforms six state-of-the-art hashing techniques based on hyperplanes across various image benchmarks of sizes ranging from one to 75 million of GIST descriptors. The performance gains are consistent and large, up to 100% improvements. The excellent results confirm the unique merits of the proposed idea in using hyperspheres to encode proximity regions in high-dimensional spaces. Finally, our method is intuitive and easy to implement.

Solid state drives (SSDs) are emerging as an alternative storage medium to HDDs. SSDs have performance characteristics (e.g., fast random reads) that are very different from those of HDDs. Because of the high performance of SSDs, there are increasingly more research efforts to redesign the established techniques that are optimized for HDDs, to work well with SSDs. In this paper we focus on computing cache-coherent layouts of large-scale models for SSDs. It has been demonstrated that cache-oblivious layouts perform well for various applications running on HDDs. However, computing cache-oblivious layouts for large-models is known to be very expensive. Also these layouts cannot be maintained efficiently for dynamically changing models. Utilizing the properties of SSDs we propose an efficient layout computation method that produces a page-based cache-aware layout for SSDs. We show that the performance of our layout can be maintained under dynamic changes on the model and is similar to the cache-oblivious layout optimized for static models. We demonstrate the benefits of our method for large-scale walkthrough scene editing and rendering, and collision detection.

We present a novel culling algorithm to perform fast and robust continuous collision detection between deforming volume meshes. This includes a continuous separating axis test that can conservatively check whether two volume meshes overlap during a given time interval. In addition, we present efficient methods to eliminate redundant elementary tests between the features (e.g., vertices, edges, and faces) of volume elements (e.g., tetrahedra, hexahedra, triangular prisms, etc.). Our approach is applicable to various deforming meshes, including those with changing topologies, and efficiently computes the first time of contact. We are able to perform inter-object and intra-object collision queries in models represented with tens of thousands of volume elements at interactive rates on a single CPU core. Moreover, we observe more than an order of magnitude performance improvement over prior methods.

Abstract We present a novel, hybrid parallel continuous collision detection (HPCCD) method that exploits the availability of multi‐core CPU and GPU architectures. HPCCD is based on a bounding volume hierarchy (BVH) and selectively performs lazy reconstructions. Our method works with a wide variety of deforming models and supports self‐collision detection. HPCCD takes advantage of hybrid multi‐core architectures – using the general‐purpose CPUs to perform the BVH traversal and culling while GPUs are used to perform elementary tests that reduce to solving cubic equations. We propose a novel task decomposition method that leads to a lock‐free parallel algorithm in the main loop of our BVH‐based collision detection to create a highly scalable algorithm. By exploiting the availability of hybrid, multi‐core CPU and GPU architectures, our proposed method achieves more than an order of magnitude improvement in performance using four CPU‐cores and two GPUs, compared to using a single CPU‐core. This improvement results in an interactive performance, up to 148 fps, for various deforming benchmarks consisting of tens or hundreds of thousand triangles.