Self Correspondence Distillation for End-to-End Weakly-Supervised Semantic Segmentation
Authors: Rongtao Xu, Changwei Wang, Jiaxi Sun, Shibiao Xu, Weiliang Meng, Xiaopeng Zhang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the PASCAL VOC 2012 and MS COCO 2014 datasets demonstrate that our method significantly outperforms other state-of-the-art methods. |
| Researcher Affiliation | Academia | 1NLPR, Institute of Automation, Chinese Academy of Sciences 2School of Artificial Intelligence, Beijing University of Posts and Telecommunications 3School of Artificial Intelligence, University of Chinese Academy of Sciences |
| Pseudocode | No | The paper describes methods using mathematical formulas and diagrams but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github.com/Rongtao Xu/Representation Learning/tree/main/SCD-AAAI2023. |
| Open Datasets | Yes | We evaluate our method on the commonly used PASCAL VOC 2012 (Everingham et al. 2010) dataset and MS COCO 2014 (Lin et al. 2014) dataset. |
| Dataset Splits | Yes | The PASCAL VOC 2012 dataset consists of training, validation and test sets with a total of 21 semantic classes. Usually the SBD dataset (Hariharan et al. 2011) is used for the augmentation of the PASCAL VOC 2012 dataset, and the augmented dataset includes 10,582 images for training, 1,449 images for validation, and 1,464 images for testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU model, CPU type, memory size) used for running the experiments. |
| Software Dependencies | No | The paper mentions implementing the method using 'PyTorch framework' but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | For the training phase, we employ the Adam W optimizer with an initial learning rate set to 6 10 5 and a weight decay factor 0.01. We use simple data augmentation strategies such as random rescaling, random horizontal flipping, and random cropping. We set the batch size to 8 and the crop size to 512 512 . For VARM, we set (α, β) to (4, 0.01). Besides, we follow (Ru et al. 2022) and set the dilation rate of the dilated convolution to [1, 2, 4, 8, 12, 24]. The number of samples n for SCD loss is set to 40. For pseudo-label generation, we follow (Ru et al. 2022) and set the two background thresholds to 0.55 and 0.35, respectively. For the experiments on PASCAL VOC 2012, the total number of iterations is set to 20,000, with 2,000 iterations warmed up for the classification branch. For the experiments of MS COCO 2014, the total number of iterations is set to 80,000, with 5,000 iterations of warm-up. |