Aug 2026· International Conference on Multimedia Analysis and Pattern Recognition· pp. 256-261· 0 citations· 15 references
Abstract
Amodal appearance completion in open-world scenarios requires reconstructing the hidden regions of occluded objects, a task that demands both geometric extrapolation and high-level semantic understanding. Conventional methods predominantly rely on geometric priors, often failing to maintain semantic coherence and structural integrity in complex scenes. In this paper, we propose Semantic Critic Amodal (SCAmodal), a novel reasoning-aware framework designed for category-agnostic amodal completion. By integrating the high-level semantic reasoning of Vision-Language Models (VLMs) with the geometric structural priors of diffusion models, SCAmodal effectively bridges the gap between scene understanding and pixel-level synthesis. The core innovation of our framework is an iterative Semantic Critic mechanism that the VLM acts as a closed-loop validator to enforce texture continuity and structural plausibility. Extensive evaluations on the challenging COCO-A and LAION benchmarks demonstrate that SCAmodal significantly outperforms existing methods, achieving superior visual realism, structural preservation, and perceptual consistency in restoring severely occluded objects.
This paper investigates amodal mask completion under a prior-guided setting, isolating shape reconstruction from upstream detection errors using oracle visible masks and semantic labels. We propose a lightweight 5-channel Swin-UNet that fuses RGB, visible, and edge priors, optimized via a novel Occlusion-Aware Loss to...
V. Nguyen, N. Phan, Tien-Dung Mai· International Conference on...· 0 citations
GaussVLA is proposed, a Mamba-based VLA that incorporates two custom modules: Gaussian Spatial Tokenizer (GST) to lift frozen semantic and depth features into compact 3D Gaussian tokens, and Depth-Aware Chain-of-Thought (DA-CoT) that performs structured, non-autoregressive geometric reasoning under language and flow-ti...
MD SELIM SAROWAR, Md Tanvir Islam, Sungho Kim et al.· 1 citation
CausalSplat is a framework that integrates vision-language models with 3D scene graphs to disentangle explicit structural perception from implicit logical inference and achieves state of the art performance on reasoning benchmarks while showing strong generalizability on standard referring and open vocabulary 3D segmen...
Jiayu Ding, Meilu Song, Yun Chen et al.· 0 citations
FactoSR is presented, a factorized reinforcement learning framework that explicitly interpret the dimensions collapsed by visual projection, and suggests that reinforcing explicit, factorized 4D consistency is a critical step toward evolving VLMs into robust, world-aware reasoners.
Yi-Jun Yang, Shenghe Zheng, Wen-Bo Li et al.· 0 citations
Space Tokens is introduced, a lightweight, architecture-agnostic framework that equips VLMs with explicit continuous spatial representations without requiring additional inference-time modules, and demonstrates that continuous spatial tokens provide an effective, interpretable, and computationally efficient mechanism f...
Hunter Schofield, Mohammed Elmahgiubi, Mohammad Mahdavian et al.· 0 citations
Large self-supervised vision models learn representations that support scene segmentation and the semantic decomposition of physical objects. We ask whether their representational geometry supports transfer to visual reasoning problems without any task-specific fine-tuning. We hypothesized that relational representatio...
Shahar Haim, Daniel C. McNamee· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.