Flow-based Generative Models for Learning Manifold to Manifold Mappings

Authors: Xingjian Zhen, Rudrasis Chakraborty, Liu Yang, Vikas Singh11042-11052

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For experiments, on a large dataset from the Human Connectome Project (HCP), we show promising results where we can reliably and accurately reconstruct brain images of a field of orientation distribution functions (ODF) from diffusion tensor images (DTI)...
Researcher Affiliation Academia 1 University of Wisconsin Madison 2 University of California, Berkeley
Pseudocode No The paper provides mathematical definitions and functional descriptions of layers (e.g., in Table 1 and Table 2) but does not include a clearly labeled "Pseudocode" or "Algorithm" block.
Open Source Code Yes *Code and supplementary materials are available at https://github.com/zhenxingjian/Dual_Manifold_GLOW
Open Datasets Yes The earth texture images dataset was introduced in (Yu et al. 2019).
Dataset Splits No The paper specifies training and test set sizes for both datasets (e.g., "The train (and test) set have 896 (and 98) images" for texture images, and "852 were used as training and 213 as the test set" for HCP) but does not explicitly mention or quantify a validation set split.
Hardware Specification No The paper discusses memory requirements and mentions GPUs generally ("entire 3D models for brain images are still too large to fit into the GPU memory") but does not provide specific details on the hardware (e.g., exact GPU/CPU models, memory amounts, or specific computing environments) used for running the experiments.
Software Dependencies No The paper mentions using "FSL s eddy" and the "Diffusion Imaging in Python (DIPY) toolbox" but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes We use 6 basic blocks of our manifold GLOW, and after every 2 blocks, reduce the resolution by half. This setup is the same for both DTI and ODF. We use 3 residual network blocks to map the latent space from DTI to ODF.