Code for paper "Neural-Sim: Learning to Generate Training Data with NeRF"

Code for paper "Neural-Sim: Learning to Generate Training Data with NeRF"
Abstract: Training computer vision models usually requires collecting and labeling vast amounts of imagery under a diverse set of scene configurations and properties. This process is incredibly time-consuming, and it is challenging to ensure that the captured data distribution maps well to the target domain of an application scenario. Recently, synthetic data has emerged as a way to address both of these issues. However, existing approaches either require human experts to manually tune each scene property or use automatic methods that provide little to no control; this requires rendering large amounts of random data variations, which is slow and is often suboptimal for the target domain. We present the first fully differentiable synthetic data pipeline that uses Neural Radiance Fields (NeRFs) in a closed-loop with a target application's loss function. Our approach generates data on-demand, with no human labor, to maximize accuracy for a target task. We illustrate the effectiveness of our method on synthetic and real-world object detection tasks. We also introduce a new "YCB-in-the-Wild" dataset and benchmark that provides a test scenario for object detection with varied poses in real-world environments.

Neural-Sim: Learning to Generate Training Data with NeRF

[ECCV 2022] Neural-Sim: Learning to Generate Training Data with NeRF

Code are actively updating, thanks!


The code is for On-demand synthetic data generation: Given a target task and a test dataset, our approach “Neural-sim” generates data on-demand using a fully differentiable synthetic data generation pipeline which maximises accuracy for the target task.


Neural-Sim pipeline: Our pipeline finds the optimal parameters for generating views from a trained neural renderer (NeRF) to use as training data for object detection. The objective is to find the optimal NeRF rendering parameters ψ that can generate synthetic training data Dtrain, such that the model (RetinaNet, in our experiments) trained on Dtrain, maximizes accuracy on a downstream task represented by the validation set Dval


1 Installation

Start by cloning the repo:

git clone

1 install the requirement of nerf-pytorch

pip install -r requirements.txt

2 install detectorn2

2 NeRF models and dataset

Quick start

For quick start, you could download our pretrained NeRF models and created sample dataset with BlenderProc here. Then unzip it and place in .logs. (Note: if not download automatically, please right click, copy the link and open in a new tab.)

Train your own NeRF model with BlenderProc

(1) Generate Bop format images with BlenderProc

  • Follow the Installation instruction of BlenderProc

  • Download BOP dataset object toolkit used in BlenderProc. For instance, to download the YCB-V dataset toolkit, please download the "Base archive" and "Object models", two zip files. Then unzip get the ycbv folder, unzip get the models folder, move the models folder into ycbv folder. The path may look like this:


Note: It would be better to create a new virtual environment for Blenderproc synthesis.

example command

python examples/camera_sampling/config.yaml /PATH/OF/BOP/ ycbv /PATH/OF/BOP/bop_toolkit/ OUTPUT/PATH

(2) Process synthesized images to be admitted by nerf (OPENCV --> OPENGL)

if use BlenderProc synthesized image, please use


if use LatentFusion read BOP format data, please use


(3) You could train NeRF with instructions NeRF-pytorch

3 Neural_Sim Bilelve optimization pipeline

cd ./optimization

Please use the to run the end-to-end pipeline. E.g.,

python --config ../configs/nerf_param_ycbv_general.txt --object_id 2 --expname exp_ycb_synthetic --psi_pose_cats_mode 5 --test_distribution 'one_1'

'--config' indicates the NeRF parameter

'--object_id' indicates the optimized ycbv object id, here is cheese box

'--expname' indicates the name of experiment

'--psi_pose_cats_mode' indicates the bin number of starting pose distribution during training

'--test_distribution' indicates the bin number of test pose distribution

Contact / Cite

If you use (part of) our code or find our work helpful, please consider citing

  title={Neural-Sim: Learning to Generate Training Data with NeRF},
  author={Ge, Yunhao and Behl, Harkirat and Xu, Jiashu and Gunasekar, Suriya and Joshi, Neel and Song, Yale and Wang, Xin and Itti, Laurent and Vineet, Vibhav},
  journal={arXiv preprint arXiv:2207.11368},

Download Source Code

Download ZIP

Paper Preview

Aug 16, 2022