Research map / Detection and segmentation / CNN features
Calibrated Adversarial Refinement for Stochastic Semantic Segmentation
Elias Kassapis, Georgi Dikov, Deepak K. Gupta, Cedric Nugteren
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace calibration loss L_cal with cross entropy L_ce in the refinement network (ablation) | CAR · LIDC | GED (16 samples): 0.639 -> 0.264 | Improved |
| replace calibration loss L_cal with cross entropy L_ce in the refinement network (ablation) | CAR · LIDC | HM-IoU (16 samples): 0.477 -> 0.592 | Improved |
| replace calibration loss L_cal with cross entropy L_ce in the refinement network (ablation) | CAR · Modified Cityscapes | GED (16 samples): 0.632 -> 0.164 | Improved |
| increase samples used for L_cal from 1 to 20 (ablation) | CAR · LIDC | GED (16 samples): 0.644 -> 0.264 | Improved |
| increase samples used for L_cal from 5 to 20 (ablation) | CAR · LIDC | GED (16 samples): 0.278 -> 0.264 | Improved |
| increase samples used for L_cal from 1 to 20 (ablation) | CAR · LIDC | HM-IoU (16 samples): 0.494 -> 0.592 | Improved |
| add the calibration loss to the regression GAN (ablation) | GAN · 1D bimodal regression | data log-likelihood | Improved |
| replace KL term with L_cal in cVAE-GAN (beta=0) (ablation) | cVAE-GAN · LIDC | GED (16 samples): 0.577 -> 0.272 | Improved |
| replace KL term with L_cal in cVAE-GAN (beta=0) versus beta=10 (ablation) | cVAE-GAN · LIDC | GED (16 samples): 0.609 -> 0.272 | Improved |
| replace KL term with L_cal in cVAE-GAN (beta=0) (ablation) | cVAE-GAN · LIDC | HM-IoU (16 samples): 0.484 -> 0.593 | Improved |
| use probabilistic encoder (cVAE-GAN+L_cal) instead of fixed noise (CAR) (ablation) | CAR · LIDC | GED (16 samples): 0.264 -> 0.272 | No clear effect |
| condition refinement network on ground-truth distribution instead of learned calibration network (ablation) | CAR · Modified Cityscapes | GED (16 samples): 0.164 -> 0.038 | Improved |
Closest papers by meaning
- Reliability in Semantic Segmentation: Are We on the Right Track? (CVPR 2023 · similarity 0.90)
- On Calibrating Semantic Segmentation Models: Analyses and an Algorithm (CVPR 2023 · similarity 0.89)
- Transferring and Regularizing Prediction for Semantic Segmentation (CVPR 2020 · similarity 0.88)
- Category-Level Adversarial Adaptation for Semantic Segmentation using Purified Features (TPAMI 2021 · similarity 0.88)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
