Large Occluded Human Image Completion via Image-Prior Cooperating

Authors: Hengrun Zhao, Yu Zeng, Huchuan Lu, Lijun Wang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments, we demonstrate our method s superior performance compared to state-of-the-art methods. ... Experiments Datasets ... Implementation Details ... Ablation Study
Researcher Affiliation Academia Hengrun Zhao*1, Yu Zeng*2, Huchuan Lu 1, Lijun Wang1, 1Dalian University of Technology 2Johns Hopkins University
Pseudocode No The paper describes its methods in text and with architectural diagrams (e.g., Fig. 4, Fig. 6), but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/Zhao Hengrun/LOHC.
Open Datasets Yes As a result, we opted to use the AHP dataset (Zhou et al. 2021)
Dataset Splits Yes For testing, we used the original dataset s validation set, which amounted to 3400 pieces, while for training, we utilized the original dataset s training set of 53199 images.
Hardware Specification Yes We use the Pytorch framework for our implementation and train on an Nvidia A100 GPU.
Software Dependencies No The paper mentions 'Pytorch framework' and 'U-Net' but does not specify their version numbers or any other software dependencies with specific versions.
Experiment Setup Yes We trained the coarse network with a batch size of 128 and set the learning rate to 1e-4. For the refinement network, we used a smaller batch size of 16 and set the learning rate to 1e-3. The learning rates of all discriminators are set to 1e-6.