Models for car quality estimation in the https://github.com/FeMa42/quality_assessment_library repository. It is a simplyfied form of the DamianBoborzi/CarQualityClassifier. While the CarQualityClassifier models process 4 views as a sequence this model uses 1 view for classifying the model quality. This way this model can be used independent on the amount of views you want to evaluate.
Quality Assessment Library
This library provides a set of functions for assessing the quality of images and 3D objects. It is focused on estimating the quality of 3D objects using images. It provides two main functionalities:
- General metrics: metrics that can be applied to any 3D object. These metrics include: MSE, CLIP-S, Spectral_MSE, D_lambda, ERGAS, PSNR, RASE, RMSE_wind, SAM, MS-SSIM, SSIM, UQI, VIF, LPIPS, SCC, FID, IS, KID. And are mainly based on torchmetrics.
- Car quality metrics: metrics that are specifically designed for cars using SigLIP and DINOv2 embeddings as features and a classifier trained on a dataset of quality-annotated cars (https://huggingface.co/datasets/DamianBoborzi/CarQualityDataset). The metrics include a quality score (the classifier score for a high quality car), the uncertainty of the quality score, a combined score that takes both the quality and the uncertainty into account, and metrics to assess the distribution of the quality scores.
Installation
To install the dependencies, run the following command:
pip install -r requirements_min.txt
Usage of general metrics
Example usage of the metrics library can be found in metrics.ipynb.
Usage of car quality metrics
Download the model from Huggingface and put them in the models folder. If you want to use the combined model also load the PCA model from the same repository. We provide an example on how to load the models in the car_quality_metrics.ipynb notebook. Example usage of the car quality metrics can also be found in car_quality_metrics.ipynb.
Further resources
- Collection of 3D Quality Assessment Repositories: 3DQA Databases
- Aesthetic Predictor used by TRELLIS: improved-aesthetic-predictor
- Dataset and Classifier for classifying the quality of 3D vehicles: MeshFleet