Coarse-to-Fine Sparse Transformer for Hyperspectral Image Reconstruction (2203.04845v3)

Abstract: Many algorithms have been developed to solve the inverse problem of coded aperture snapshot spectral imaging (CASSI), i.e., recovering the 3D hyperspectral images (HSIs) from a 2D compressive measurement. In recent years, learning-based methods have demonstrated promising performance and dominated the mainstream research direction. However, existing CNN-based methods show limitations in capturing long-range dependencies and non-local self-similarity. Previous Transformer-based methods densely sample tokens, some of which are uninformative, and calculate the multi-head self-attention (MSA) between some tokens that are unrelated in content. This does not fit the spatially sparse nature of HSI signals and limits the model scalability. In this paper, we propose a novel Transformer-based method, coarse-to-fine sparse Transformer (CST), firstly embedding HSI sparsity into deep learning for HSI reconstruction. In particular, CST uses our proposed spectra-aware screening mechanism (SASM) for coarse patch selecting. Then the selected patches are fed into our customized spectra-aggregation hashing multi-head self-attention (SAH-MSA) for fine pixel clustering and self-similarity capturing. Comprehensive experiments show that our CST significantly outperforms state-of-the-art methods while requiring cheaper computational costs. The code and models will be released at https://github.com/caiyuanhao1998/MST

• The paper presents a novel coarse-to-fine sparse transformer that enhances hyperspectral image reconstruction by focusing on important spectral-spatial details.
• It implements sparse attention mechanisms to significantly reduce computational overhead while maintaining high-quality image restoration.
• Experimental results demonstrate notable improvements over traditional methods, suggesting promising applications in remote sensing and advanced imaging.

Overview of Submission Guidelines for ECCV

This document serves as an exemplar for authors preparing submissions to the European Conference on Computer Vision (ECCV). It outlines the expected formatting and procedural guidelines, ensuring consistency and standardization across all submissions.

Key Components of the Manuscript Guidance

The manuscript should adhere to strict formatting rules, aimed at creating a uniform look and feel. The document stresses that authors should use the provided \LaTeX{} templates and avoid altering the defined formatting guidelines. The essence is to produce a manuscript that is visually consistent, easily readable, and compliant with the ECCV standards.

Initial Submission Requirements

Submissions must be written in English and conform to a strict page limit of 14 pages, excluding references. This constraint is rigorously enforced; exceeding this limit results in outright rejection without review. Authors are advised not to manipulate margins or font sizes to bypass this restriction.

Each submission must include a unique paper ID prominently on each page to aid in the review process. Additionally, all submission lines must be numbered, assisting reviewers in efficiently providing feedback.

Review and Confidentiality Policies

ECCV's review process is double-blind. Reviewers and area chairs do not know the authors' identities and vice versa. This anonymity extends to citation practices and supplementary materials, which must remain devoid of identifying information. Adherence to this policy is crucial; violations may lead to manuscript rejection.

The confidentiality of the review process is underscored, highlighting volunteer reviewers' roles and emphasizing ethical conduct limitations.

Dual Submission and Publication Policies

The paper must not have been previously published or submitted concurrently to another peer-reviewed venue. ECCV defines "publication" to exclude non-peer-reviewed preprints and reports. The aim is to introduce genuinely novel research to the conference audience while avoiding reviewer redundancy.

Successful papers require at least one author to register and present their work at the conference, ensuring both theoretical and applied contributions are effectively communicated to the ECCV community.

Manuscript Preparation Guidelines

Authors are encouraged to use \LaTeX{} for manuscript preparation due to its efficiency in maintaining formatting standards. Guidelines detail specific instructions for typography, paper layout, and length requirements. All graphics, figures, and tables must be electronically produced and adhere to resolutions that guarantee clarity and readability.

Compliance and Submission Packaging

Upon acceptance, papers require conversion to camera-ready format. This step includes adding author names, affiliations, and acknowledgments, along with ensuring all supplementary files and signed copyright forms are submitted in the correct format.

The submission package, crucially, must include all source files and must be verified for completeness before uploading. Any discrepancies or omissions in the submission package can impact its successful integration into the proceedings.

Conclusion

In summary, this document meticulously details the procedural and technical requirements for submitting a paper to ECCV. Compliance ensures the integrity of the review process and aids in maintaining the high standards of publication. Researchers intending to contribute should thoroughly familiarize themselves with these guidelines to enhance their submission's prospects of favorable review and acceptance. Future implications suggest continued refinement of these processes to adapt to evolving publishing and technological standards.