Reproduce “Object-driven Text-to-Image Synthesis via Adversarial Training” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenbo Li, Pengchuan Zhang, Lei Zhang, Qiuyuan Huang, Xiaodong He, Siwei Lyu, Jianfeng Gao
Obj-GAN predicts layouts of object categories, bounding boxes, and shapes from captions, then uses each object’s class label to select relevant caption words during image refinement. A Fast R-CNN discriminator evaluates object regions for realism and text consistency. On COCO, predicted-layout Obj-GAN-SOTA reports an Inception score of 30.29 ± 0.33 versus 23.79 ± 0.32 for AttnGAN, but training and evaluation differ, and the comparison does not establish photorealism.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.