AdaptIS: Adaptive Instance Selection Network (1909.07829v1)

Abstract: We present Adaptive Instance Selection network architecture for class-agnostic instance segmentation. Given an input image and a point $(x, y)$, it generates a mask for the object located at $(x, y)$. The network adapts to the input point with a help of AdaIN layers, thus producing different masks for different objects on the same image. AdaptIS generates pixel-accurate object masks, therefore it accurately segments objects of complex shape or severely occluded ones. AdaptIS can be easily combined with standard semantic segmentation pipeline to perform panoptic segmentation. To illustrate the idea, we perform experiments on a challenging toy problem with difficult occlusions. Then we extensively evaluate the method on panoptic segmentation benchmarks. We obtain state-of-the-art results on Cityscapes and Mapillary even without pretraining on COCO, and show competitive results on a challenging COCO dataset. The source code of the method and the trained models are available at https://github.com/saic-vul/adaptis.

Guidelines for Submitting Manuscripts to ICCV Proceedings

The presented document serves as a comprehensive guideline for authors preparing manuscripts for the International Conference on Computer Vision (ICCV). This style guide is crucial for maintaining consistency and quality across submissions, enforcing adherence to specified formatting and submission rules.

Key Elements of the Guidelines

The document outlines several critical aspects that authors must consider when preparing their submissions:

1. Language and Length: Submissions must be in English and are limited to eight pages, not counting references. Overlength submissions are not reviewed, ensuring concise presentations of research findings.
2. Anonymity and Blind Review: The guidelines reinforce the importance of anonymizing submissions for blind review, specifying that self-citations should not suggest the author’s identity. This is crucial for maintaining objectivity in the review process.
3. Formatting Details: Strict formatting rules are provided, including two-column text layout, specific font styles (primarily Times), and layout margins. These standards are set to ensure readability and uniformity.
4. Figures and Mathematics: Authors are directed on the inclusion of equations and figures, emphasizing the need for clear labeling and appropriate sizing, which maintains visual clarity without detracting from the textual narrative.
5. Additional Materials: The document permits the inclusion of supplementary materials, such as technical reports, provided they are referenced correctly and do not distract from the main text’s self-sufficiency.
6. Final Submission Requirements: Authors are reminded to include signed IEEE copyright release forms for their papers to be eligible for publication. This legal formalization is necessary to protect intellectual property and publication rights.

Practical and Theoretical Implications

Adhering to these guidelines is essential for authors to streamline the review process and enhance the dissemination of their research at ICCV. By enforcing strict standards:

• Consistency in Proceedings: The uniform submission style ensures that conference proceedings are professionally presented, aiding readers in navigating complex technical content.
• Objective Peer Review: Anonymization contributes to a fair review process, fostering an impartial evaluation of works based solely on merit rather than author identity.
• Enhanced Clarity and Accessibility: Clear formatting and precise guidelines ensure that complex technical information is accessible, improving communication among researchers.

Speculation on Future Developments

While this document pertains specifically to ICCV 2019, evolving practices in scholarly communication may influence future iterations of the guidelines. Potential developments could include:

• Integration of Interactive Elements: As digital dissemination grows, future guidelines might accommodate interactive figures or datasets to enhance reader engagement.
• Increased Emphasis on Open Access: With the global push towards open science, there may be shifts towards more accessible publication practices, which could be reflected in new submission requirements.

This document acts as a critical resource for researchers aiming to contribute effectively to ICCV, underscoring the importance of meticulous adherence to academic standards in the field of computer vision.