add badges from paperswithcode

README.md

@@ -13,6 +13,18 @@ NeurIPS 2021 <br />
 [![Paper](https://img.shields.io/badge/arXiv-2106.11562-brightgreen)](https://arxiv.org/abs/2106.11562)
 <img src = "https://github.com/clovaai/SSUL/blob/main/figures/SSUL_main.png" width="100%" height="100%">
 
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-10-1-on-pascal-voc-2012)](https://paperswithcode.com/sota/overlapped-10-1-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-15-1-on-pascal-voc-2012)](https://paperswithcode.com/sota/overlapped-15-1-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-15-5-on-pascal-voc-2012)](https://paperswithcode.com/sota/overlapped-15-5-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-19-1-on-pascal-voc-2012)](https://paperswithcode.com/sota/overlapped-19-1-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/disjoint-10-1-on-pascal-voc-2012)](https://paperswithcode.com/sota/disjoint-10-1-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/disjoint-15-1-on-pascal-voc-2012)](https://paperswithcode.com/sota/disjoint-15-1-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/disjoint-15-5-on-pascal-voc-2012)](https://paperswithcode.com/sota/disjoint-15-5-on-pascal-voc-2012?p=ssul-semantic-segmentation-with-unknown-label)
+
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-100-5-on-ade20k)](https://paperswithcode.com/sota/overlapped-100-5-on-ade20k?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-100-50-on-ade20k)](https://paperswithcode.com/sota/overlapped-100-50-on-ade20k?p=ssul-semantic-segmentation-with-unknown-label)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/ssul-semantic-segmentation-with-unknown-label/overlapped-50-50-on-ade20k)](https://paperswithcode.com/sota/overlapped-50-50-on-ade20k?p=ssul-semantic-segmentation-with-unknown-label)
+
 # Abtract
 This paper introduces a solid state-of-the-art baseline for a class-incremental semantic segmentation (CISS) problem. While the recent CISS algorithms utilize variants of the knowledge distillation (KD) technique to tackle the problem, they failed to fully address the critical challenges in CISS causing the catastrophic forgetting; the semantic drift of the background class and the multi-label prediction issue. To better address these challenges, we propose a new method, dubbed **SSUL-M (Semantic Segmentation with Unknown Label with Memory)**, by carefully combining techniques tailored for semantic segmentation. Specifically, we claim three main contributions. (1) defining unknown classes within the background class to help to learn future classes (help plasticity), (2) freezing backbone network and past classifiers with binary cross-entropy loss and pseudo-labeling to overcome catastrophic forgetting (help stability), and (3) utilizing tiny exemplar memory for the first time in CISS to improve both plasticity and stability. The extensively conducted experiments show the effectiveness of our method, achieving significantly better performance than the recent state-of-the-art baselines on the standard benchmark datasets. Furthermore, we justify our contributions with thorough ablation analyses and discuss different natures of the CISS problem compared to the traditional class-incremental learning targeting classification.