ARC: Anchored Representation Clouds for High-Resolution INR Classification

2025-03-19

AI Summary

Existing implicit neural representation (INR)-based image classification methods suffer from sensitivity to spatial transformations and insufficient local modeling capability, leading to degraded performance on high-resolution images. To address these limitations, we propose Spatially Anchored Latent Cloud (SALC), the first INR framework to explicitly incorporate image coordinates as geometric anchors for latent variables—thereby endowing the representation with explicit positional awareness. SALC mitigates translation non-robustness and high-frequency detail loss inherent in global MLP encoding by jointly integrating coordinate embedding, localized latent cloud modeling, and a lightweight decoder architecture. Experiments demonstrate that SALC achieves state-of-the-art performance on both low- and high-resolution INR classification benchmarks. It significantly improves translation robustness while maintaining memory efficiency and strong generalization across diverse domains.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Representation Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Implicit neural representations (INRs) encode signals in neural network weights as a memory-efficient representation, decoupling sampling resolution from the associated resource costs. Current INR image classification methods are demonstrated on low-resolution data and are sensitive to image-space transformations. We attribute these issues to the global, fully-connected MLP neural network architecture encoding of current INRs, which lack mechanisms for local representation: MLPs are sensitive to absolute image location and struggle with high-frequency details. We propose ARC: Anchored Representation Clouds, a novel INR architecture that explicitly anchors latent vectors locally in image-space. By introducing spatial structure to the latent vectors, ARC captures local image data which in our testing leads to state-of-the-art implicit image classification of both low- and high-resolution images and increased robustness against image-space translation. Code can be found at https://github.com/JLuij/anchored_representation_clouds.

Problem

• Improves high-resolution image classification using INRs.
• Addresses sensitivity to image-space transformations in INRs.
• Introduces local representation to enhance detail and robustness.

Innovation

• Local anchoring of latent vectors in image-space
• Enhanced high-resolution image classification accuracy
• Improved robustness against image-space transformations