metadata

language:
  - en
license: other
license_name: autodesk-non-commercial-3d-generative-v1.0
tags:
  - wala
  - single-view-to-3d

Model Card for WaLa-SV-1B

This model is part of the Wavelet Latent Diffusion (WaLa) paper, capable of generating high-quality 3D shapes from single-view images with detailed geometry and complex structures.

Model Details

Model Description

WaLa-SV-1B is a large-scale 3D generative model trained on a massive dataset of over 10 million publicly-available 3D shapes. It can efficiently generate a wide range of high-quality 3D shapes from single-view image inputs in just 2.5 seconds. The model uses a wavelet-based compact latent encoding and a billion-parameter architecture to achieve superior performance in terms of geometric detail and structural plausibility.

Developed by: Aditya Sanghi, Aliasghar Khani, Chinthala Pradyumna Reddy, Arianna Rampini, Derek Cheung, Kamal Rahimi Malekshan, Kanika Madan, Hooman Shayani
Model type: 3D Generative Model
License: Autodesk Non-Commercial (3D Generative) v1.0

For more information please look at the Project Page and the paper.

Model Sources

Repository: Github
Paper: ArXiv:TBD
Demo: TBD

Uses

Direct Use

This model is released by Autodesk and intended for academic and research purposes only for the theoretical exploration and demonstration of the WaLa 3D generative framework. Please see here for inferencing instructions.

Out-of-Scope Use

The model should not be used for:

Commercial purposes
Creation of load-bearing physical objects the failure of which could cause property damage or personal injury
Any usage not in compliance with the license, in particular, the "Acceptable Use" section.

Bias, Risks, and Limitations

Bias

The model may inherit biases present in the publicly-available training datasets, which could lead to uneven representation of certain object types or styles.
The model's performance may degrade for object categories or styles that are underrepresented in the training data.

Risks and Limitations

The quality of the generated 3D output may be impacted by the quality and clarity of the input.
The model may occasionally generate implausible shapes, especially when the input is ambiguous or of low quality. Even theoretically plausible shapes should not be relied upon for real-world structural soundness.

How to Get Started with the Model

Please refer to the instructions here

Training Details

Training Data

The model was trained on a dataset of over 10 million 3D shapes aggregated from 19 different publicly-available sub-datasets, including ModelNet, ShapeNet, SMLP, Thingi10K, SMAL, COMA, House3D, ABC, Fusion 360, 3D-FUTURE, BuildingNet, DeformingThings4D, FG3D, Toys4K, ABO, Infinigen, Objaverse, and two subsets of ObjaverseXL (Thingiverse and GitHub).

Training Procedure

Preprocessing

Each 3D shape in the dataset was converted into a truncated signed distance function (TSDF) with a resolution of 256³. The TSDF was then decomposed using a discrete wavelet transform to create the wavelet-tree representation used by the model.

Training Hyperparameters

Training regime: Please refer to the paper.

Speeds, Sizes, Times

The model contains approximately 956 million parameters.
The model can generate shapes within 2-4 seconds.

Evaluation

Testing Data, Factors & Metrics

Testing Data

The model was evaluated on the Google Scanned Objects (GSO) dataset and a validation set from the training data (MAS validation data).

Factors

The evaluation considered various factors such as the quality of generated shapes, the ability to capture fine details and complex structures, and the model's performance across different object categories.

Metrics

The model was evaluated using the following metrics:

Intersection over Union (IoU)
Light Field Distance (LFD)
Chamfer Distance (CD)

Results

The single-view to 3D model achieved the following results on the GSO dataset:

LFD: 2509.20
IoU: 0.6154
CD: 0.02150

On the MAS validation dataset:

LFD: 2920.74
IoU: 0.6056
CD: 0.01530

Technical Specifications

Model Architecture and Objective

TThe model uses a U-ViT architecture with modifications. It employs a wavelet-based compact latent encoding to effectively capture both coarse and fine details of 3D shapes from single-view inputs. The input view is processed through the DINO v2 encoder to extract feature representations, which then serve as the condition latent vectors for the generative model.

ADSKAILab
/

WaLa-SV-1B