Unsupervised Model Adaptation for Continual Semantic Segmentation

Authors: Serban Stan, Mohammad Rostami2593-2601

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on benchmark adaptation tasks demonstrate our method achieves competitive performance even compared with joint UDA approaches. Finally, we provide experiments on the GTA5 Cityscapes and SYNTHIA Cityscapes benchmark domain adaptation image segmentation tasks to demonstrate that our method is effective and leads to competitive performance, even when compared against existing UDA algorithms.
Researcher Affiliation Academia 1 University of Southern California 2 Information Sciences Institute sstan@usc.edu, rostamim@usc.edu
Pseudocode Yes Algorithm 1 MAS3 (λ, τ)
Open Source Code Yes We validate our algorithm using two benchmark domain adaptation tasks, and provide our code at https://github.com/serbanstan/mas3-continual.
Open Datasets Yes We validate MAS3 on the standard GTA5 (Richter et al. 2016) Cityscapes (Cordts et al. 2016) and the SYNTHIA (Ros et al. 2016) Cityscapes benchmark UDA tasks for semantic segmentation.
Dataset Splits Yes Cityscapes is a real-world dataset consisting of a training set with 2,957 instances and a validation set, used as testing set, with 500 instances of images with size 2040 1016.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud computing specifications used for running its experiments.
Software Dependencies No The paper mentions using 'Deep Lab V3' and 'VGG16' as models but does not provide specific version numbers for software dependencies such as deep learning frameworks (e.g., PyTorch, TensorFlow) or their required versions.
Experiment Setup No Due to space limits, implementation details are included in the Appendix. This indicates that specific experimental setup details are not present in the main text.