Fader Networks:Manipulating Images by Sliding Attributes

Authors: Guillaume Lample, Neil Zeghidour, Nicolas Usunier, Antoine Bordes, Ludovic DENOYER, Marc'Aurelio Ranzato

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments, 5.1 Experiments on the celeb A dataset, Quantitative evaluation protocol, Quantitative results, 5.2 Experiments on Flowers dataset. We performed a quantitative evaluation of Fader Networks on Mechanical Turk, using Ic GAN as a baseline.
Researcher Affiliation Collaboration 1Facebook AI Research 2Sorbonne Universités, UPMC Univ Paris 06, UMR 7606, LIP6 3LSCP, ENS, EHESS, CNRS, PSL Research University, INRIA
Pseudocode No The paper describes the architecture and algorithm in text and mathematical equations but does not provide structured pseudocode or algorithm blocks.
Open Source Code No The paper refers to a GitHub link for the Ic GAN baseline model (https://github.com/Guim3/Ic GAN) but does not provide a link or statement about open-sourcing the Fader Networks code itself.
Open Datasets Yes We first present experiments on the celeb A dataset [14], which contains 200, 000 images of celebrity of shape 178 × 218 annotated with 40 attributes. We performed additional experiments on the Oxford-102 dataset, which contains about 9, 000 images of flowers classified into 102 categories [17].
Dataset Splits Yes We used the standard training, validation and test split.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments (e.g., GPU/CPU models, memory, or cloud instance types).
Software Dependencies No The paper mentions "All models were trained with Adam [11]" but does not provide specific software dependencies with version numbers (e.g., Python, TensorFlow, PyTorch versions) needed for replication.
Experiment Setup Yes All models were trained with Adam [11], using a learning rate of 0.002, β1 = 0.5, and a batch size of 32. We performed data augmentation by flipping horizontally images with a probability 0.5 at each iteration. We initially set λE to 0 and the model is trained like a normal auto-encoder. Then, λE is linearly increased to 0.0001 over the first 500, 000 iterations to slowly encourage the model to produce invariant representations.