A Layer-Based Sequential Framework for Scene Generation with GANs

Authors: Mehmet Ozgur Turkoglu, William Thong, Luuk Spreeuwers, Berkay Kicanaoglu8901-8908

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Via quantitative and qualitative experiments on a subset of the MS-COCO dataset, we show that our proposed framework produces not only more diverse images but also copes better with affine transformations and occlusion artifacts of foreground objects than its counterparts.
Researcher Affiliation Academia Mehmet Ozgur Turkoglu,1 William Thong,2 Luuk Spreeuwers,1 Berkay Kicanaoglu2 1University of Twente, 2University of Amsterdam
Pseudocode No The paper describes the model architecture and processes using text and figures, but no structured pseudocode or algorithm blocks are provided.
Open Source Code Yes The code is available at https: //github.com/0zgur0/Seq Scene Gen.
Open Datasets Yes MS-COCO dataset (Lin et al. 2014) is used to evaluate the performance of our proposed model.
Dataset Splits Yes The dataset contains 164K training images over 80 semantic classes. Additionally, we also compute the mean Io U scores on the images generated conditioned on semantic maps of the validation set (450 images).
Hardware Specification No No specific hardware details (like GPU/CPU models or memory) are provided for running the experiments.
Software Dependencies No The paper mentions optimizers and network architectures but does not specify software dependencies like programming languages, libraries, or frameworks with version numbers (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes Parameters are updated with the Adam optimizer (β1 = 0, β2 = 0.9, learning rate of 2e 4 and divided by 2 every 80 epochs) (Kingma and Ba 2014). All the models are trained for 480 epochs with a batch size of 16. The parameters of the generators are updated after 5 updates of the discriminator. The tradeoff hyper-parameters in the foreground generator loss function (Eq. 8) are set to λl = 0.1, λr = 1e 5, λfm = 1 and in the background generator loss function (Eq. 3) to λr = 100, λfm = 1.