ReGIR: Refine Geometry for Sigle-Image Implicit Representation (ICASSP2024)

Contribution:

A novel query point encoding utilizing pixel-aligned feature, voxel-aligned feature, and SDF.
coarse-to-fine implicit model based on multi-level voxel-aligned feature for 3D human reconstruction.

This project is built upon ICON, and the core idea of the work is inspired by PIFuHD.

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
apps		apps
assets		assets
configs		configs
docs		docs
examples_image		examples_image
lib		lib
sample_data/thuman2		sample_data/thuman2
scripts		scripts
LICENSE		LICENSE
ReadME.md		ReadME.md
colab.sh		colab.sh
command.txt		command.txt
environment.yaml		environment.yaml
fetch_cape.sh		fetch_cape.sh
fetch_data.sh		fetch_data.sh
fetch_hps.sh		fetch_hps.sh
requirements.txt		requirements.txt
requirements_colab.txt		requirements_colab.txt