Learn, Imagine and Create: Text-to-Image Generation from Prior Knowledge
Authors: Tingting Qiao, Jing Zhang, Duanqing Xu, Dacheng Tao
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Thorough experiments on two public benchmark datasets demonstrate Leica GAN s superiority over the baseline method. |
| Researcher Affiliation | Academia | Tingting Qiao1,2 Jing Zhang2 Duanqing Xu1 Dacheng Tao2 1College of Computer Science and Technology, Zhejiang University, China 2UBTECH Sydney AI Centre, School of Computer Science, Faculty of Engineering The University of Sydney, Darlington, NSW 2008, Australia |
| Pseudocode | No | The paper describes the model architecture and mathematical formulations but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code has been made available at https://github.com/qiaott/Leica GAN. |
| Open Datasets | Yes | We evaluated our model on two commonly used datasets, i.e. the CUB bird [33] and Oxford-102 flower [16]. |
| Dataset Splits | No | The paper explicitly states the size of the training and testing sets for CUB and Oxford-102 datasets but does not provide specific details for a validation set split. 'CUB containing 8,855 training and 2,933 testing data belonging to 200 categories, and Oxford containing 7,034 training and 1,155 testing data belonging to 102 categories.' |
| Hardware Specification | No | No specific hardware details (e.g., CPU/GPU models, memory, or cloud instance types) used for running experiments are provided in the paper. |
| Software Dependencies | No | The paper mentions using specific models like bi-directional LSTM and Inception-v3 and refers to libraries via citations but does not specify software dependencies with version numbers (e.g., PyTorch version, TensorFlow version, CUDA version). |
| Experiment Setup | Yes | The dimension D was 256, the sentence length O was 18 and the image region size H was 299 299. α1, α2, α3 and λw were set to 4, 5, 10, 1. The balance weights δI = 0.8 and δM = 0.2. The weights for training TVE of Leica GAN with the best performance on the CUB bird dataset were γ1 = 1, γ2 = 1, γ3 = 4, γ4 = 1, γ5 = 1,γ6 = 0.5. On the Oxford-102 flower dataset, the best weights were γ1 = 1, γ2 = 1, γ3 = 0, γ4 = 1, γ5 = 0.5, γ6 = 0. |