Reducing Information Bottleneck for Weakly Supervised Semantic Segmentation
Authors: Jungbeom Lee, Jooyoung Choi, Jisoo Mok, Sungroh Yoon
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental evaluations demonstrate that this simple modification significantly improves the quality of localization maps on both the PASCAL VOC 2012 and MS COCO 2014 datasets, exhibiting a new state-of-the-art performance for weakly supervised semantic segmentation. The code is available at: https://github.com/jbeomlee93/RIB. |
| Researcher Affiliation | Academia | Jungbeom Lee1 Jooyoung Choi1 Jisoo Mok1 Sungroh Yoon1,2, 1 Department of Electrical and Computer Engineering, Seoul National University 2 Interdisciplinary Program in Artificial Intelligence, Seoul National University |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found. |
| Open Source Code | Yes | The code is available at: https://github.com/jbeomlee93/RIB. |
| Open Datasets | Yes | We evaluated our method quantitatively and qualitatively by conducting experiments on the PASCAL VOC 2012 [16] and the MS COCO 2014 [39] datasets. |
| Dataset Splits | Yes | We used the PASCAL VOC 2012 dataset, which is augmented by Hariharan et al. [20], containing 10,582 training images with objects from 20 classes. The MS COCO 2014 dataset contains approximately 82K training images containing objects of 80 classes. We evaluated our method on 1,449 validation images and 1,456 test images from the PASCAL VOC 2012 dataset and on 40,504 validation images from the MS COCO 2014 dataset |
| Hardware Specification | No | The paper mentions models like ResNet-50 and Deep Lab-v2-ResNet101, and frameworks like PyTorch, but does not provide specific details on the hardware (e.g., GPU models, CPU types) used for running the experiments in the main text. It refers to the Appendix for this information, which is not provided. |
| Software Dependencies | No | The paper mentions 'Py Torch framework [43]' but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We fine-tuned our classifier for K = 10 iterations with a learning rate of 8 10 6 and a batch size of B = 20. We set the margin m to 600. For the GNDRP, we set τ to 0.4. For the MS COCO 2014 dataset, we cropped the training images with the crop size of 481 481 rather than 321 321 used for the PASCAL VOC 2012 dataset, considering the size of the images in this dataset. |