Skip to yearly menu bar Skip to main content


[Re] Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation

Aryan Mehta · Karan Uppal · Kaushal Jadhav · Monish Natarajan · Mradul Agrawal · Debashish Chakravarty

Keywords: [ ReScience - MLRC 2021 ] [ Journal Track ]


The following submission is a reproducibility report for 'Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation' published in CVPR 2021 as part of the ML Reproducibility Challenge 2021. The paper’s central claim revolves around the newly introduced Background Aware Pooling method to generate high-quality pseudo labels using bounding boxes as supervision and Noise Aware Loss to train a segmentation network using those noisy labels. We started with the publicly available code-base provided by the authors and reproduced the results involving pseudo label generation. Further, we implemented Noise-Aware Loss and used it to train a semantic segmentation network, reproducing its claims. We performed many refactoring and upgrades on the author's code to include various procedures mentioned in the paper as well as reimplemented the code base in PyTorch Lightning for easier reproducibility. To formulate the report, extensive experimentation on the author's code has been conducted to verify all the claims made by the author described in detail below. We also performed additional experiments to gauge the extent of improvement brought upon by the method over the state-of-the-art methods.

Chat is not available.