Towards Scalable Unpaired Virtual Try-On via Patch-Routed Spatially-Adaptive GAN

Authors: Zhenyu Xie, Zaiyu Huang, Fuwei Zhao, Haoye Dong, Michael Kampffmeyer, Xiaodan Liang

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Experiments, We conduct experiments on two existing virtual try-on benchmark datasets (MPV [6] dataset and Deep Fashion [21] dataset) and our newly collected large-scale benchmark dataset for unpaired try-on, named UPT.
Researcher Affiliation Collaboration 1Shenzhen Campus of Sun Yat-Sen University, 2Huya Inc 3Ui T The Arctic University of Norway, 4Peng Cheng Laboratory
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No Our code will be available at PASTA-GAN.
Open Datasets Yes We conduct experiments on two existing virtual try-on benchmark datasets (i.e., the Deep Fashion [21] and the MPV [6] datasets) and our newly collected large-scale benchmark dataset for unpaired try-on, named UPT.
Dataset Splits No The paper states for the UPT dataset: "UPT is further split into a training set of 27,139 images and a testing set of 6,115 images." but does not explicitly mention a validation set split. For other datasets, general training and testing sets are mentioned without explicit split percentages or counts for all three stages.
Hardware Specification Yes Our PASTA-GAN is implemented using Py Torch [28] and is trained on 8 Tesla V100 GPUs.
Software Dependencies No The paper states: "Our PASTA-GAN is implemented using Py Torch [28]" but does not provide a specific version number for PyTorch or any other software dependencies.
Experiment Setup Yes During training, the batch size is set to 96 and the model is trained for 4 million iterations with a learning rate of 0.002 using the Adam optimizer [16] with β1 = 0 and β2 = 0.99. The loss hyper-parameters λrec, λperc, and λmask are set to 40, 40, and 100, respectively. The hhyper-parameters for the random erasing probability α1 and α2 are set to 0.2 and 0.9, respectively.