One Weird Trick to Improve Your Semi-Weakly Supervised Semantic Segmentation Model

Authors: Wonho Bae, Junhyug Noh, Milad Jalali Asadabadi, Danica J. Sutherland

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate prediction filtering on PASCAL VOC 2012 [Everingham et al., 2015] which contains 10,582 training, 1,449 validation, and 1,456 test images. For SWSSS, we follow the training splits of Ouali et al. [2020], where 1,464 images are used for Dpixel. As with previous work, we evaluate segmentation performance by mean Intersection over Union (m Io U), generally on the validation set. Test set performance is obtained from the PASCAL VOC evaluation server, without any tricks such as multi-scale or flipping.
Researcher Affiliation Collaboration Wonho Bae1 , Junhyug Noh2 , Milad Jalali Asadabadi1 and Danica J. Sutherland1,3 1University of British Columbia 2Lawrence Livermore National Laboratory 3Alberta Machine Intelligence Institute
Pseudocode Yes We provide full pseudocode in the appendix.
Open Source Code No The paper mentions 'See the supplementary material' but does not explicitly state that the source code for the methodology described is publicly available or provide a link to a repository.
Open Datasets Yes We evaluate prediction filtering on PASCAL VOC 2012 [Everingham et al., 2015] which contains 10,582 training, 1,449 validation, and 1,456 test images.
Dataset Splits Yes We evaluate prediction filtering on PASCAL VOC 2012 [Everingham et al., 2015] which contains 10,582 training, 1,449 validation, and 1,456 test images. For SWSSS, we follow the training splits of Ouali et al. [2020], where 1,464 images are used for Dpixel.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper mentions various model architectures like Deeplab V1, Deeplab V3, OCRNet, and ResNet50, but does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup No The paper states that prediction filtering introduces 'a single additional hyperparameter, the threshold τ', but it does not specify the value of this threshold or other specific hyperparameters like learning rates, batch sizes, or optimizer settings for their experiments within the main text.