Practical overview
When it fits
Use MICA when a research or development workflow needs metrically scaled 3D face geometry reconstructed from images, or when working with its documented training and evaluation scripts.
Who it is for
- Computer-vision researchers working on 3D face reconstruction
- Developers who need image-to-mesh outputs and FLAME parameters
- Researchers preparing MICA training data or running the provided evaluation workflows
Documented uses
- Convert face images into .ply meshes, rendered images and .npy FLAME parameters.
- Prepare ArcFace input images for MICA training using the repository's dataset scripts.
- Run reconstruction evaluation with the provided testing scripts and supported evaluation datasets.
Requirements & limits
The README documents a Conda environment and requires obtaining FLAME2020 through a FLAME account. Its install script requests FLAME account credentials to access the model files.
- FLAME2020 and pretrained MICA and InsightFace model files are part of the documented setup.
- The demo expects input images in demo/input/ and creates per-subject outputs.
- Training requires images from all eight datasets described by the repository; training from scratch with ArcFace initialization also requires Glint360K.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up MICA for the image demo
Prepare the documented environment and model files needed to use the demo.
Before you start
- A Conda environment
- A FLAME account to obtain the FLAME2020 model
- The pretrained MICA model and InsightFace antelopev2 and buffalo_l models
Steps
Clone the MICA repository and enter its directory.
Create the environment from environment.yml, or use the repository's install.sh script.
Obtain FLAME2020 through a FLAME account and place it in /data/pretrained/FLAME2020/; the install script requests FLAME account credentials to access files.
Ensure the pretrained MICA and InsightFace models are in the locations specified by the README; the install script handles their placement.
Expected outcome
The documented environment and model files are prepared for the demo workflow.
Guide limitations
- The README's install instructions describe setup but do not specify hardware requirements.
- The FLAME2020 model requires access through a FLAME account.
Generate face-reconstruction outputs with the demo
Process the demo input images into the repository's documented per-subject outputs.
Before you start
- The MICA repository environment and required model files are prepared.
- Images are available in demo/input/.
Steps
Place the images to process in demo/input/.
Run demo.py.
Expected outcome
The demo creates an output destination for each subject containing a .ply mesh, a rendered image and .npy FLAME parameters.
Guide limitations
- The README does not specify supported image formats or hardware requirements.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- Zielon/MICA · Official homepage title · retained in its source language