Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RankED: Addressing Imbalance and Uncertainty in Edge Detection Using Ranking-based Losses

Bedrettin Cetinkaya, Sinan Kalkan, Emre Akbas

TL;DR

RankED is proposed, a unified ranking-based approach that addresses both the imbalance problem (P 1) and the uncertainty problem (P2) and tackles these two problems with two components: One component which ranks positive pixels over negative pixels, and the second which promotes high confidence edge pixels to have more label certainty.

Abstract

Detecting edges in images suffers from the problems of (P1) heavy imbalance between positive and negative classes as well as (P2) label uncertainty owing to disagreement between different annotators. Existing solutions address P1 using class-balanced cross-entropy loss and dice loss and P2 by only predicting edges agreed upon by most annotators. In this paper, we propose RankED, a unified ranking-based approach that addresses both the imbalance problem (P1) and the uncertainty problem (P2). RankED tackles these two problems with two components: One component which ranks positive pixels over negative pixels, and the second which promotes high confidence edge pixels to have more label certainty. We show that RankED outperforms previous studies and sets a new state-of-the-art on NYUD-v2, BSDS500 and Multi-cue datasets. Code is available at https://ranked-cvpr24.github.io.

RankED: Addressing Imbalance and Uncertainty in Edge Detection Using Ranking-based Losses

TL;DR

RankED is proposed, a unified ranking-based approach that addresses both the imbalance problem (P 1) and the uncertainty problem (P2) and tackles these two problems with two components: One component which ranks positive pixels over negative pixels, and the second which promotes high confidence edge pixels to have more label certainty.

Abstract

Detecting edges in images suffers from the problems of (P1) heavy imbalance between positive and negative classes as well as (P2) label uncertainty owing to disagreement between different annotators. Existing solutions address P1 using class-balanced cross-entropy loss and dice loss and P2 by only predicting edges agreed upon by most annotators. In this paper, we propose RankED, a unified ranking-based approach that addresses both the imbalance problem (P1) and the uncertainty problem (P2). RankED tackles these two problems with two components: One component which ranks positive pixels over negative pixels, and the second which promotes high confidence edge pixels to have more label certainty. We show that RankED outperforms previous studies and sets a new state-of-the-art on NYUD-v2, BSDS500 and Multi-cue datasets. Code is available at https://ranked-cvpr24.github.io.
Paper Structure (16 sections, 15 equations, 3 figures, 9 tables, 1 algorithm)

This paper contains 16 sections, 15 equations, 3 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology: RankED
  4. Ranking Positive Edge Pixels over Negatives
  5. Sorting Positives with their Uncertainties
  6. Computing Pixel Uncertainties
  7. Overall Loss Function
  8. Optimization
  9. Experiments
  10. Experimental Setup and Details
  11. Experiment 1: Comparison with SOTA
  12. Experiment 2: Uncertainty-aware Evaluation
  13. Experiment 3: Ablation Analysis
  14. Experiment 4: Qualitative Comparison
  15. Experiment 5: Running Time Comparisions
  16. ...and 1 more sections

Figures (3)

  • Figure 1: (a) Current approaches threshold label certainties and class-balanced cross-entropy loss for training edge detectors. (b) With RankED, we propose a unified approach which ranks positives over negatives to handle the imbalance problem and sorts positives with respect to their certainties.
  • Figure 2: An overview of RankED. RankED introduces two novel loss functions for edge detection: $\mathcal{L}_{\textrm{Rank}}$ ($\S$\ref{['sect:ranking']}) for ranking positive pixels over background pixels, and $\mathcal{L}_{\textrm{Sort}}$ ($\S$\ref{['sect:sorting']}) for sorting positive pixels with respect to their (un)certainties ($\S$\ref{['sect:uncertainty']}).
  • Figure 3: Qualitative results on BSDS dataset. All outputs are obtained after the post-processing step. Red: OIS scores.