Disentangled Embedding through Style and Mutual Information for Domain Generalization

Authors: Noaman Mehmood, Kenneth Barner

TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on the PACS, Office Home, VLCS and Terra Incognita datasets show that both frameworks outperform several state-of-the-art DG techniques.
Researcher Affiliation Academia Noaman Mehmood EMAIL Department of Electrical and Computer Engineering University of Delaware Kenneth Barner EMAIL Department of Electrical and Computer Engineering University of Delaware
Pseudocode Yes Algorithm 1 Training procedure for DETMI Algorithm 2 Training procedure for DETSI
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We conducted extensive experiments on the PACS Li et al. (2017), Office-Home Venkateswara et al. (2017), VLCS Torralba & Efros (2011), and Terra Incognita Beery et al. (2018) benchmarks.
Dataset Splits Yes PACS: This study uses the official training-validation split. VLCS: The training and validation split follows the methodology described in Li et al. (2018b). Office-Home: The training and validation split adheres to the approach outlined in Xu et al. (2021). Terra Incognita: Following the protocol in Cha et al. (2022), we use 80% of the data from the source domain for training and reserve the remaining 20% for validation.
Hardware Specification No The paper does not explicitly describe the specific hardware used to run its experiments, such as GPU models or CPU specifications.
Software Dependencies No The paper mentions using ResNet50 as a feature extractor but does not provide specific version numbers for any software libraries, frameworks, or programming languages used (e.g., PyTorch, TensorFlow, Python versions).
Experiment Setup Yes The model is trained using mini-batch stochastic gradient descent (SGD) with a batch size of 32 for the PACS, Office-Home, VLCS, and and Terra Incognita datasets. The training process spans 50 epochs, with a weight decay of 5e 4 and an initial learning rate of 0.001. The learning rate is reduced by a factor of 0.1 after every 40 epochs. The weighting coefficients λ1 and λ2 are set to 1. Standard data augmentation techniques are applied, including color jittering, horizontal flipping, and random resized cropping. All input images are resized to 224 224 to ensure consistency during training.