Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

PLOP: Learning without Forgetting for Continual Semantic Segmentation (2011.11390v3)

Published 23 Nov 2020 in cs.CV

Abstract: Deep learning approaches are nowadays ubiquitously used to tackle computer vision tasks such as semantic segmentation, requiring large datasets and substantial computational power. Continual learning for semantic segmentation (CSS) is an emerging trend that consists in updating an old model by sequentially adding new classes. However, continual learning methods are usually prone to catastrophic forgetting. This issue is further aggravated in CSS where, at each step, old classes from previous iterations are collapsed into the background. In this paper, we propose Local POD, a multi-scale pooling distillation scheme that preserves long- and short-range spatial relationships at feature level. Furthermore, we design an entropy-based pseudo-labelling of the background w.r.t. classes predicted by the old model to deal with background shift and avoid catastrophic forgetting of the old classes. Our approach, called PLOP, significantly outperforms state-of-the-art methods in existing CSS scenarios, as well as in newly proposed challenging benchmarks.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Arthur Douillard (20 papers)
  2. Yifu Chen (20 papers)
  3. Arnaud Dapogny (33 papers)
  4. Matthieu Cord (129 papers)
Citations (213)

Summary

An Examination of Author Rebuttal Guidelines for CVPR Submissions

The document titled "LaTeX Guidelines for Author Response" provides a comprehensive set of instructions for authors preparing rebuttals to reviewers' comments on papers submitted to the Conference on Computer Vision and Pattern Recognition (CVPR). These guidelines are crucial for maintaining a standardized response format and ensuring that submissions adhere to the expectations of the review process. This adherence is vital for both the integrity of the review process and the clarity of communication between authors and reviewers.

The document emphasizes that the author rebuttal is an optional component designed to facilitate discussions around factual errors identified by reviewers or to provide additional details requested by them. It is explicitly stated that the rebuttal is not intended to introduce new contributions, such as theorems or algorithms, that were absent in the original submission. The guidelines align with the 2018 PAMI-TC motion, which advises reviewers not to demand additional experiments for the rebuttal phase or penalize authors for omission thereof. This measure ensures that the rebuttal process is not used to pressure authors into conducting new, potentially time-consuming experiments that would not have been feasible within the original submission timeline.

The rebuttal must adhere to strict formatting rules, including a one-page limit, which encompasses all references and figures. Authors are required to maintain the same blind submission criteria as in the original paper submission, enforcing uniformity and fairness. The format prescribes a two-column layout with specific margin requirements, consistent with previous CVPR specifications. Such detailed formatting instructions are crucial for maintaining consistency across submissions, allowing reviewers to focus on the content rather than being distracted by formatting discrepancies.

A significant highlight of the document is the guidance on the presentation of graphics. Authors are advised to ensure that any visual data included in the rebuttal is legible when printed. This recommendation underscores the importance of accessibility and readability in academic discourse, as reviewers may choose to print documents rather than review them digitally.

The document also outlines specific typographical requirements, such as using Times or Times Roman fonts, setting main text in 10-point single-spaced font, and ensuring section headings are appropriately sized. Consistency in bibliographic references and citations is also mandated, emphasizing the need for precise and standardized scholarly communication.

In conclusion, the guidelines provided in this document have significant implications for the conduct of the CVPR rebuttal process. By setting clear expectations and limitations, these guidelines foster a fair review environment where the emphasis is on resolving factual disputes and clarifying submitted work rather than introducing new unreviewed content. As AI and computer vision continue to develop, the structure provided by such guidelines will be essential in maintaining rigorous review standards in increasingly competitive academic landscapes. Future adjustments to these guidelines may reflect evolving practices in scientific communication, but the core tenets of clarity, fairness, and consistency will likely remain unchanged.

Youtube Logo Streamline Icon: https://streamlinehq.com