Rethinking conditional GAN training: An approach using geometrically structured latent manifolds

Authors: Sameera Ramasinghe, Moshiur Farazi, Salman H Khan, Nick Barnes, Stephen Gould

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

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the efficacy of our solution on a baseline c GAN (i.e., Pix2Pix) which lacks diversity, and show that by only modifying its training mechanism (i.e., with our proposed Pix2Pix-Geo), one can achieve more diverse and realistic outputs on a broad set of image-to-image translation tasks. ... In this section, we demonstrate the effectiveness of the proposed training scheme using qualitative and quantitative experiments. ... Table 1 depicts the quantitative results. As shown, our model exhibits a higher diversity and a higher realism on multiple datasets. In all the cases, we outperform our baseline by a significant margin. ... We conduct an ablation study to compare the different variants of the proposed technique. Table 2 depicts the results.
Researcher Affiliation Collaboration Sameera Ramasinghe The Australian National University, Data61-CSIRO sameera.ramasinghe@anu.edu.au Moshiur Farazi Data61-CSIRO Salman Khan Mohamed Bin Zayed University of AI Nick Barnes The Australian National University Stephen Gould The Australian National University
Pseudocode Yes Algorithm 1: Training algorithm
Open Source Code Yes Code available at: https://github.com/samgregoost/Rethinking-CGANs
Open Datasets Yes We validate the efficacy of our solution on a baseline c GAN (i.e., Pix2Pix) ... on a broad set of image-to-image translation tasks. ... We compare our method against state-of-the-art models that focus on multimodal image-to-image translation. Fig. 2 shows the qualitative results on landmarks faces, sketch anime and BW color. Table 1 depicts the quantitative results on 9 (nine) challenging datasets (e.g., facades2photo, sat2map, edges2shoes, edges2bags, sketch2anime, BW2color, lm2faces, hog2faces, night2day).
Dataset Splits No The paper does not explicitly specify training, validation, and test dataset splits with percentages or sample counts. While it mentions 'train' in Algorithm 1, it does not provide explicit details for a validation split.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments, such as specific GPU or CPU models, or cloud computing resources.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as 'Python 3.8' or 'PyTorch 1.9'.
Experiment Setup No The paper states 'For further details on the datasets and hyper-parameter settings, see App. I.', indicating that specific experimental setup details are provided in an appendix rather than in the main text.