Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Creative Sketch Generation
Authors: Songwei Ge, Vedanuj Goswami, Larry Zitnick, Devi Parikh
ICLR 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Quantitative evaluations as well as human studies demonstrate that sketches generated by our approach are more creative and of higher quality than existing approaches. |
| Researcher Affiliation | Collaboration | Songwei Ge University of Maryland, College Park EMAIL Vedanuj Goswami & C. Lawrence Zitnick Facebook AI Research EMAIL Devi Parikh Facebook AI Research Georgia Institute of Technology EMAIL |
| Pseudocode | No | No explicit pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | Yes | Our datasets, code, and a web demo are publicly available 1. songweige.github.io/projects/creative_sketech_generation/home.html |
| Open Datasets | Yes | Our datasets, code, and a web demo are publicly available 1. songweige.github.io/projects/creative_sketech_generation/home.html ... To this end, we trained an Inception model on the Quick Draw3.8M dataset (Xu et al., 2020). |
| Dataset Splits | Yes | The dataset contains 345 classes and each class contains 9, 000 training samples, 1, 000 validation samples, and 1, 000 test samples. |
| Hardware Specification | Yes | Our training time of each creature and bird part generator is approximately 4 and 2 days on a single NVIDIA Quadro GV100 Volta GPU. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and Style GAN2 architecture but does not specify version numbers for general software dependencies like Python, PyTorch/TensorFlow, or CUDA. |
| Experiment Setup | Yes | We picked a learning rate of 10^-4 and a batch size of 40 for both the discriminator and generator. We use the Adam optimizer... with β1 = 0, β2 = 0.99, ϵ = 10^-8. ... We train the creature part generators for 60, 000 steps and bird part generators for 30, 000 steps. |