Generating Interpretable Images with Controllable Structure

By Scott E. Reed, Aäron van den Oord, Nal Kalchbrenner, Victor Bapst, Matthew Botvinick, Nando de Freitas. 2017

65 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Image-to-Image Translation with Conditional Adversarial Networks (2016, background): “[44] for an autoregressive approach to the same problem).”
StackGAN: Text to Photo-Realistic Image Synthesis with Stacked Generative Adversarial Networks (2016, methodology): “[27] used conditional PixelCNN to generate images using the text descriptions and object location constraints.”
Parallel Multiscale Autoregressive Density Estimation (2017, methodology): “These networks have been used in a wide variety of conditional image generation schemes such as text and spatial structure to image (Mansimov et al., 2015; Reed et al., 2016b;a; Wang & Gupta, 2016).”
Generating Multiple Objects at Spatially Distinct Locations (2019, methodology): “…a randomly sampled noise vector z, the location and size of the individual bounding boxes bboxi, a label for each of the bounding boxes encoded as a one-hot vector lonehoti , and, if existent, an image caption embedding ϕ obtained with a pretrained char-CNN-RNN network from Reed et al. (2016b).”
Learning Visual Reasoning Without Strong Priors (2017, background): “Other approaches [27, 28, 29, 30, 31, 32, 33] employ a similar, repetitive conditioning, so perhaps there is an underlying principle that explains the success of these approaches.”
CPGAN: Content-Parsing Generative Adversarial Networks for Text-to-Image Synthesis (2020, methodology): “Text-to-image synthesis was initially investigated based on pixelCNN [35, 37], which suffers from highly computational cost during the inference phase.”
Disentangled Person Image Generation (2017, methodology): “in [26] proposed the Generative Adversarial What-Where Network that uses pose keypoints and text descriptions as condition, whereas in [27] they used an extension of PixelCNN in addition to conditioning on part keypoints, segmentation masks and text to generate images on the MPI Human Pose dataset, ”
CPGAN: Full-Spectrum Content-Parsing Generative Adversarial Networks for Text-to-Image Synthesis (2019, methodology): “Text-to-image synthesis was initially investigated based on pixelCNN [24, 25], which suffers from highly computational cost during the inference phase.”
Adversarial Text-to-Image Synthesis: A Review (2021, background): “The follow-up work [105] extends PixelCNN [106] to generate images from captions with controllable object locations leveraging keypoints and masks.”
CKD: Cross-Task Knowledge Distillation for Text-to-Image Synthesis (2020, background): “To re-lieve this issue, some researchers [10], [13], [14] introduce their frameworks with constraints about global structures or local details.”
Semantic Object Accuracy for Generative Text-to-Image Synthesis (2019, background): “[27] and [28] build on [10] by additionally conditioning the generator on bounding boxes or keypoints of relevant objects.”