qimaqi / Implicit-Zoo

Official Implementation of Impliti-Zoo: Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes

Implicit-Zoo 🦜: A Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes

Introduction | Download Link | Installation | Quick Start | Results Demo | News |

Introduction

This repository contains the download link, example code, test results for the paper Implicit-Zoo: A Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes. It showcase the effectiveness of optimizing monocular camera poses as a continuous function of time with neural network.

We have released the demo code, more details will be released soon, please check news for details.

Download Link

Here we provide download link for CIFAR-10-INRS, ImageNet-100-INRs and Omniobject3D from Kaggle. For data size reason you can find ImageNet-1K-INRs in google drive link. For CityScapes-INRs the cityscapes we will actively discuss this detail with the Cityscapes team and provide an update as soon as possible.

Installation

conda env create --file environment.yml
conda activate implicit_zoo

Quick Start

An introduction notebook for Dataset visualization:

Generate CIFAR-DATA

bash cifar_main_exps.sh

Note that you can customize config in experiments/cifar_generate_configs/main.yaml Like customize network depth and width or training iteration times. Moreover the default CIFAR data installed place is in ./data. You can also change in code experiments/generate_cifar_dataset_siren.py line 54.

Results Demo

Time Cost

Visualize of queried Images

Diagram of learnable token

CIFAR-10 Exps results

News

• Create the repo
• upload CIFAR-10 Dataset

• upload ImageNet-100 Dataset
• upload ImageNet-1k Dataset
• upload Omniobject3D Dataset
• upload notebook demo
• upload reproduce code

Citation

@misc{ma2024implicitzoolargescaledatasetneural,
      title={Implicit-Zoo: A Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes}, 
      author={Qi Ma and Danda Pani Paudel and Ender Konukoglu and Luc Van Gool},
      year={2024},
      eprint={2406.17438},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2406.17438}, 
}