WACV 2026 Papers

Gaussian Splatting has proven to be an effective algorithm for novel view synthesis and 3D reconstruction from multi-view images. However, the underlying volumetric primitive -- the ellipsoidal Gaussian-- has limited expressive capabilities, leading to difficulties in 3D modelling (especially geometry such as edges, corners, and high curvature). To address this limitation, in this paper, we introduce Superquadric Splats (SQS), an extended class of volumetric primitives, as a super-set of Gaussian splats, to model more detailed geometry. We treat superquadrics as volumetric distance functions rather than level-set surfaces. A non-trivial differentiable rendering pipeline is developed to support this. Extensive experimental analysis on multiple datasets validates the effectiveness of the proposed SQS approach, showing both enhanced visual and geometric performance compared to Gaussian-based splatting (with more than 1dB in PSNR and prominent geometric improvement).

Occlusion Boundary Estimation (OBE) identifies boundaries arising from both inter-object occlusions and self-occlusion within individual objects. This task is closely related to Monocular Depth Estimation (MDE), which infers depth from a single image, as Occlusion Boundaries (OBs) provide critical geometric cues for resolving depth ambiguities, while depth can conversely refine occlusion reasoning. In this paper, we aim to systematically model and exploit this mutually beneficial relationship. To this end, we propose MoDOT, a novel framework for joint estimation of depth and OBs, which incorporates a new Cross-Attention Strip Module (CASM) to leverage mid-level OB features for depth prediction, and a novel OB-Depth Constraint Loss (OBDCL) to enforce geometric consistency. To facilitate this study, we contribute OB-Hypersim, a large-scale photorealistic dataset with precise depth and self-occlusion-handled OB annotations. Extensive experiments on two synthetic datasets and NYUD-v2 demonstrate that MoDOT achieves significantly better performance than single-task baselines and multi-task competitors. Furthermore, models trained solely on our synthetic data demonstrate strong generalization to real-world scenes without fine-tuning, producing depth maps with sharper boundaries and improved geometric fidelity. Collectively, these results underscore the significant benefits of jointly modeling OBs and depth. Code and resources are available at https://github.com/xul-ops/MoDOT.

Machine learning is evolving towards high-order models that necessitate pre-training on extensive datasets, a process associated with significant overheads. Traditional models, despite having pre-trained weights, are becoming obsolete due to architectural differences that obstruct the effective transfer and initialization of these weights. To address these challenges, we introduce a novel framework, QuadraNet V2, which leverages quadratic neural networks to create efficient and sustainable high-order learning models. Our method initializes the primary term of the quadratic neuron using a standard neural network, while the quadratic term is employed to adaptively enhance the learning of data non-linearity or shifts. This integration of pre-trained primary terms with quadratic terms, which possess advanced modeling capabilities, significantly augments the information characterization capacity of the high-order network. By utilizing existing pre-trained weights, QuadraNet V2 reduces the required GPU hours for training by 90% to 98.4% compared to training from scratch, demonstrating both efficiency and effectiveness.

Image inpainting restores missing regions in a visually plausible way, yet existing methods struggle with irregular holes and complex structures due to fixed kernels and static attention. In this paper, we propose Mask-Aware Deformable Inpainting Network (MADIN), which incorporates mask information for position-aware control. The proposed model employs a Two-branch Offset Estimator that jointly utilizes query features and mask signals to reliably predict reference positions even within masked regions. In addition, Adaptive Offset Range Scaling adjusts offset magnitude based on masking ratio to capture broader context when needed. Experiments on CelebA-HQ and Places2 show that MADIN achieves state-of-the-art results in PSNR, SSIM, LPIPS, and FID, while remaining lightweight and efficient. The code will be released at https://github.com/ohhhyeongsn/MADIN

We present Large Sign Language Models (LSLM), a novel framework for translating 3D American Sign Language (ASL) by leveraging Large Language Models (LLMs) as the backbone, which can benefit hearing-impaired individuals' virtual communication. Unlike existing sign language recognition methods that rely on 2D video, our approach directly utilizes 3D sign language data to capture rich spatial, gestural, and depth information in 3D scenes. This enables more accurate and resilient translation, enhancing digital communication accessibility for the hearing-impaired community. Beyond the task of ASL translation, our work explores the integration of complex, embodied multimodal languages into the processing capabilities of LLMs, moving beyond purely text-based inputs to broaden their understanding of human communication. We investigate both direct translation from 3D gesture features to text and an instruction-guided setting where translations can be modulated by external prompts, offering greater flexibility. This work provides a foundational step toward inclusive, multimodal intelligent systems capable of understanding diverse forms of language.

Variational Autoencoders (VAEs) form the foundation of modern video generation models. In particular, discrete latent VAEs with vector quantization have gained prominence for their superior perceptual sharpness, ability to model long-range dynamics, and efficient adaptation to downstream tasks. However, existing discrete VAEs face two key limitations: (i) a lack of frequency-domain modeling to enhance global spatiotemporal understanding, and (ii) fixed-resolution quantization schemes, preventing effective modeling of coarse-to-fine spatiotemporal hierarchies essential for video generation. To address these limitations, we propose a Pyramidal Vector Quantized Variational Autoencoder (PVQ-VAE) for videos. PVQ-VAE's encoder-decoder leverages Fast Fourier Transform and Discrete Wavelet Transform to capture global semantics and multi-scale local details jointly. We introduce Pyramidal Vector Quantization (PVQ), a hierarchical quantization scheme that discretizes features at multiple resolutions to better capture multi-scale information. To further boost fidelity, we introduce a cross-modal contrastive loss guided by a pretrained high-resolution image VAE. PVQ-VAE achieves state-of-the-art performance on WebVid-val, COCO-val, and MCL-JCV, reconstructing videos with high perceptual quality at up to 32x spatial and 16x temporal compression.

Despite significant advances in diffusion models, achieving precise and composable image editing without task-specific training remains a challenge. Existing approaches often rely on iterative optimization or linear latent operations, which are slow, brittle, and prone to attribute entanglement (e.g., editing "lipstick" inadvertently alters skin tone). We introduce SphereEdit, a training-free framework that leverages the spherical geometry of diffusion embeddings and token aware cross-attention to enable interpretable, fine-grained control. We represent semantic attributes as unit vector directions in the denoiser's prediction space and show that antipodal symmetry ("old" is approximately the negation of "young") naturally supports bidirectional edits, while approximate orthogonality enables clean composition through spherical coefficient. At inference, these directions modulate cross-attention activations, producing spatially localized edits without optimization or fine-tuning. SphereEdit achieves sharper, more disentangled edits than prior baselines, while remaining plug-and-play and applicable across diverse image editing tasks.

Not identical but similar objects are ubiquitous in our world, ranging from four-legged animals such as dogs and cats to cars of different models and flowers of various colors. This study addresses a novel task of matching such non-identical objects at the pixel level. We propose a weighting scheme of descriptors, Semantic Enhancement Weighting (SEW), that incorporates semantic information from object detectors into existing sparse feature matching methods, extending their targets from identical objects captured from different perspectives to semantically similar objects. The experiments show successful matching between non-identical objects in various cases, including in-class design variations, class discrepancy, and domain shifts (e.g., photo vs. drawing and image corruptions). The code is available at https://github.com/Circ-Leaf/NIOM.

Establishing object correspondence between egocentric (ego) and exocentric (exo) views is a critical capability for robot learning and human-robot interaction. The core task involves segmenting an object in one view given a query mask from the opposing view. This is notoriously difficult due to cluttered scenes with many task-irrelevant objects and drastic appearance changes across perspectives. To address this, we introduce RegionAligner, a unified text-visual framework that strategically focuses learning on task-relevant regions. Our method first uses a large vision-language model to identify and name salient objects, effectively filtering out visual distractors. These object phrases are then fused with visual features from both views. We introduce a novel region-guided supervision strategy that promotes focus, enforces spatial alignment, and minimizes appearance disparity between the ego-exo views. Furthermore, our framework seamlessly adapts to unsupervised settings by automatically generating pseudo-labels from matched mask proposals, drastically reducing annotation costs. Extensive experiments on the challenging Ego-Exo4D dataset show RegionAligner significantly outperforms existing baselines, improving IoU by 10.16% (ego-to-exo) and 6.04% (exo-to-ego).

Recent advancements in the direct training of Spiking Neural Networks (SNNs) have demonstrated high-quality outputs even at early timesteps, paving the way for novel energy-efficient AI paradigms. However, the inherent non-linearity and temporal dependencies in SNNs introduce persistent challenges, such as temporal covariate shift (TCS) and unstable gradient flow with learnable neuron thresholds. In this paper, we present two key innovations: MP-Init (Membrane Potential Initialization) and TrSG (Threshold-robust Surrogate Gradient). MP-Init addresses TCS by aligning the initial membrane potential with its stationary distribution, while TrSG stabilizes gradient flow with respect to threshold voltage during training. Extensive experiments validate our approach, achieving state-of-the-art accuracy on both static and dynamic image datasets. The code is available at: https://github.com/kookhh0827/SNN-MP-Init-TRSG

In this work, we propose an all-in-one video restoration framework that grounds degradation-aware semantic context of video frames in natural language via foundation models, offering interpretable and flexible guidance. Unlike prior art, our method assumes no degradation knowledge in train or test time and learns an approximation to the grounded knowledge such that the foundation model can be safely disentangled during inference adding no extra cost. Further, we call for standardization of benchmarks in all-in-one video restoration, and propose two benchmarks in multi-degradation setting, three-task (3D) and four-task (4D), and two time-varying composite degradation benchmarks; one of the latter being our proposed dataset with varying snow intensity, simulating how weather degradations affect videos naturally. We compare our method with prior works and report state-of-the-art performance on all benchmarks.

The creation of 3D scenes has traditionally been both labor-intensive and costly, requiring designers to meticulously configure 3D assets and environments. Recent advancements in generative AI, including text-to-3D and image-to-3D methods, have dramatically reduced the complexity and cost of this process. However, current techniques for editing complex 3D scenes continue to rely on generally interactive multi-step, 2D-to-3D projection methods and diffusion-based techniques, which often lack precision in control and hamper interactive-rate performance. In this work, we propose ***3DSceneEditor***, a fully 3D-based paradigm for interactive-rate, precise editing of intricate 3D scenes using Gaussian Splatting. Unlike conventional methods, 3DSceneEditor operates through a streamlined 3D pipeline, enabling direct Gaussian-based manipulation for efficient, high-quality edits based on input prompts. The proposed framework (i) integrates a pre-trained instance segmentation model for semantic labeling; (ii) employs a zero-shot grounding approach with CLIP to align target objects with user prompts; and (iii) applies scene modifications, such as object addition, repositioning, recoloring, replacing, and removal--directly on Gaussians. Extensive experimental results show that 3DSceneEditor surpasses existing state-of-the-art techniques in terms of both editing precision and efficiency, establishing a new benchmark for efficient and interactive 3D scene customization.

Professional photo editing remains challenging, requiring extensive knowledge of imaging pipelines and significant expertise. While recent deep learning approaches, particularly style transfer methods, have attempted to automate this process, they often struggle with output fidelity, editing control, and complex retouching capabilities. We propose a novel retouch transfer approach that learns from professional edits through before-after image pairs, enabling precise replication of complex editing operations. We develop a context-aware Implicit Neural Representation that learns to apply edits adaptively based on image content and context, and is capable of learning from a single example. Our method extracts implicit transformations from reference edits and adaptively applies them to new images. To facilitate this research direction, we introduce a comprehensive Photo Retouching Dataset comprising 100,000 high-quality images edited using over 170 professional Adobe Lightroom presets. Through extensive evaluation, we demonstrate that our approach not only surpasses existing methods in photo retouching but also enhances performance in related image reconstruction tasks like Gamut Mapping and Raw Reconstruction. By bridging the gap between professional editing capabilities and automated solutions, our work presents a significant step toward making sophisticated photo editing more accessible while maintaining high-fidelity results.

Existing video highlight detection methods, although advanced, struggle to generalize well to all test videos. These methods typically employ a generic highlight detection model for each test video, which is suboptimal as it fails to account for the unique characteristics and variations of individual test videos. Such fixed models do not adapt to the diverse content, styles, or audio and visual qualities present in new, unseen test videos, leading to reduced highlight detection performance. In this paper, we propose Highlight-TTA, a test-time adaptation framework for video highlight detection that addresses this limitation by dynamically adapting the model during testing to better align with the specific characteristics of each test video, thereby improving generalization and highlight detection performance. Highlight-TTA is jointly optimized with an auxiliary task, cross-modality hallucinations, alongside the primary highlight detection task. We utilize a meta-auxiliary training scheme to enable effective adaptation through the auxiliary task while enhancing the primary task. During testing, we adapt the trained model using the auxiliary task on the test video to further enhance its highlight detection performance. Extensive experiments with three state-of-the-art highlight detection models and three benchmark datasets show that the introduction of Highlight-TTA to these models improves their performance, yielding superior results.

High-quality 3D streaming from multiple cameras is crucial for immersive experiences in many AR/VR applications. The limited number of views - often due to real-time constraints - leads to missing information and incomplete surfaces in the rendered images. Existing approaches typically rely on simple heuristics for the hole filling, which can result in inconsistencies or visual artifacts. We propose to complete the missing textures using a novel, application-targeted inpainting method independent of the underlying representation as an image-based post-processing step after the novel view rendering. The method is designed as a standalone module compatible with any calibrated multi-camera system. For this we introduce a multi-view aware, transformer-based network architecture using spatio-temporal embeddings to ensure consistency across frames while preserving fine details. Additionally, our resolution-independent design allows adaptation to different camera setups, while an adaptive patch selection strategy balances inference speed and quality, allowing real-time performance. We evaluate our approach against state-of-the-art inpainting techniques under the same real-time constraints and demonstrate that our model achieves the best trade-off between quality and speed, outperforming competitors in both image and video-based metrics.

Inner interpretability is a promising field aiming to uncover the internal mechanisms of AI systems through scalable, automated methods. While significant research has been conducted on large language models, limited attention has been paid to applying inner interpretability to large-scale image tasks, focusing primarily on architectural and functional levels to visualize learned concepts. In this paper, we first present a conceptual framework that supports inner interpretability and multilevel analysis for large-scale image classification tasks. Specifically, we introduce the Bi-directional Interaction between Concept and Input Embeddings (Bi-ICE) module, which facilitates interpretability across the computational, algorithmic, and implementation levels. This module enhances transparency by generating predictions based on human-understandable concepts, quantifying their contributions, and localizing them within the inputs. Finally, we showcase enhanced transparency in image classification, measuring concept contributions, and pinpointing their locations within the inputs. Our approach highlights algorithmic interpretability by demonstrating the process of concept learning and its convergence.

Multimodal Large Language Models (MLLMs) struggle with accurately capturing camera-object relations, especially for object orientation, camera viewpoint, and camera shots. This stems from the fact that existing MLLMs are trained on images with limited diverse camera-object relations and corresponding textual descriptions. To address this, we propose a synthetic generation pipeline to create large-scale 3D visual instruction datasets. Our framework takes 3D assets as input and uses rendering and diffusion-based image generation models to create photorealistic images preserving precise camera-object relations. Additionally, large language models (LLMs) are used to generate text prompts for guiding visual instruction tuning and controlling image generation. We create Ultimate3D, a dataset of **240K** VQAs with precise camera-object annotations, and corresponding benchmark. MLLMs fine-tuned on our proposed dataset outperform commercial models by a large margin, achieving an average accuracy improvement of **33.4%** on camera-object relation recognition tasks. Our code, dataset, and benchmark will contribute to broad MLLM applications.

Temporal action segmentation is a critical task in video understanding, where the goal is to assign action labels to each frame in a video. While recent advances leverage iterative refinement-based strategies, they fail to explicitly utilize the hierarchical nature of human actions. In this work, we propose HybridTAS - a novel framework that incorporates a hybrid of Euclidean and hyperbolic geometries into the denoising process of diffusion models to exploit the hierarchical structure of actions. Hyperbolic geometry naturally provides tree-like relationships between embeddings, enabling us to guide the action label denoising process in a coarse-to-fine manner: higher diffusion timesteps are influenced by abstract, high-level action categories (root nodes), while lower timesteps are refined using fine-grained action classes (leaf nodes). Extensive experiments on three benchmark datasets, GTEA, 50Salads, and Breakfast, demonstrate that our method achieves state-of-the-art performance, validating the effectiveness of hyperbolic-guided denoising for the temporal action segmentation task.

We address Embodied Reference Understanding, the task of predicting the object a person in the scene refers to through pointing gesture and language. This requires multimodal reasoning over text, visual pointing cues, and scene context, yet existing methods often fail to fully exploit visual disambiguation signals. We also observe that while the referent often aligns with the head-to-fingertip direction, in many cases it aligns more closely with the wrist-to-fingertip direction, making a single-line assumption overly limiting. To address this, we propose a dual-model framework, where one model learns from the head-to-fingertip direction and the other from the wrist-to-fingertip direction. We introduce a Gaussian ray heatmap representation of these lines and use them as input to provide a strong supervisory signal that encourages the model to better attend to pointing cues. To fuse their complementary strengths, we present the CLIP-Aware Pointing Ensemble module, which performs a hybrid ensemble guided by CLIP features. We further incorporate an auxiliary object center prediction head to enhance referent localization. We validate our approach on YouRefIt, achieving 75.0 mAP at 0.25 IoU, alongside state-of-the-art CLIP and C_D scores, and demonstrate its generality on unseen CAESAR and ISL Pointing, showing robust performance across benchmarks.