MOST-GAN: 3D Morphable StyleGAN for Disentangled Face Image Manipulation

Authors: Safa C. Medin, Bernhard Egger, Anoop Cherian, Ye Wang, Joshua B. Tenenbaum, Xiaoming Liu, Tim K. Marks1962-1971

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

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, we manipulate portrait images with respect to several physical attributes and compare them with the results of a state-of-the-art relighting method (Hou et al. 2021) and a state-of-the-art real-image manipulation method, PIE (Tewari et al. 2020b). Besides providing qualitative comparisons with these two methods, we also quantitatively compare the performance of our pose editing algorithm by employing a head pose estimator (Ruiz, Chong, and Rehg 2018) to measure the error between the desired and estimated head poses.
Researcher Affiliation Collaboration 1Mitsubishi Electric Research Laboratories (MERL) 2Massachusetts Institute of Technology 3Friedrich-Alexander-University Erlangen-Nuremberg 4Michigan State University
Pseudocode No The paper describes the architecture and process in narrative text and diagrams, but does not provide structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the methodology described.
Open Datasets Yes After pretraining, we train our model using the FFHQ face dataset (Karras, Laine, and Aila 2019)... we extract hair masks from the USC Hair Salon database (Hu et al. 2015)...
Dataset Splits No The paper mentions training and testing on datasets but does not explicitly specify detailed training/validation/test splits (e.g., percentages or counts for each split, or references to how the FFHQ dataset was split into these portions for the authors' experiments).
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper mentions using StyleGAN2 and MichiGAN but does not specify version numbers for these or any other software dependencies.
Experiment Setup No The paper describes the training process with loss functions but does not specify concrete hyperparameters like learning rate, batch size, or number of epochs for the experimental setup.