Towards High-Fidelity Face Self-Occlusion Recovery via Multi-View Residual-Based GAN Inversion

Authors: Jinsong Chen, Hu Han, Shiguang Shan294-302

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our approach outperforms the state-of-the-art methods in face selfocclusion recovery under unconstrained scenarios.
Researcher Affiliation Academia 1 Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China 2 University of Chinese Academy of Sciences, Beijing 100049, China 3 Pengcheng National Laboratory, Shenzhen 518055, China chenjinsong20@mails.ucas.ac.cn, {hanhu, sgshan}@ict.ac.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or a link to the open-source code for the described methodology.
Open Datasets Yes The face image datasets we used for training are Celeb A (Liu et al. 2018) and FFHQ dataset collected by (Karras, Laine, and Aila 2019).
Dataset Splits No The paper mentions using Celeb A and FFHQ datasets for training and MOFA-test for evaluation, but does not provide specific details on train/validation/test splits (e.g., percentages, sample counts, or explicit standard split references).
Hardware Specification Yes We implement our method with torch(1.7.1) and Py Torch3D (v0.4.0), and run our experiments on NVIDIA 1080Ti GPUs with Intel 2.1GHz CPUs.
Software Dependencies Yes We implement our method with torch(1.7.1) and Py Torch3D (v0.4.0)
Experiment Setup Yes We set hyper-parameters λrec = 1.9, λperc = 0.2 following (Deng et al. 2019b), and the other hyper-parameters empirically: λid = 0.8, λadv = 0.1. We set the input image size to 224 × 224 and the number of vertices and triangle faces to 35, 709 and 70, 897 respectively, the same as (Shang et al. 2020).