jneuendorf/rendr

---
license: mit
task_categories:
- image-classification
- zero-shot-image-classification
pretty_name: RENDR
size_categories:
- 10K<n<100K
tags:
- 3d
- synthetic
- rendered
- computer-graphics
dataset_info:
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': animals
'1': appliances
'2': architecture
'3': decoration
'4': electronics
'5': furniture
'6': lighting
'7': mechanical
'8': nature
'9': people
'10': tools
splits:
- name: train
num_bytes: 1459477777.312
num_examples: 23836
- name: test
num_bytes: 236313059.742
num_examples: 4206
download_size: 1584949874
dataset_size: 1695790837.0540001
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# RENDR Dataset

## Dataset Description

RENDR is a large-scale synthetic dataset of rendered 3D objects across 11 object categories. The dataset contains rendered images from 3D assets sourced from BlenderKit and Haven, designed for training and evaluating computer vision models on synthetic 3D data.

## Dataset Statistics

### Split Overview

| Split | Total Images | Rendered | BlenderKit Assets | Haven Assets |
|-------|--------------|----------|-------------------|--------------|
| Train | 29,291 | 23,836 | 5,397 | 58 |
| Test | 5,929 | 4,206 | 1,701 | 22 |

### Class Distribution

| Class | Train (Rendered) | Train (BlenderKit) | Train (Haven) | Test (Rendered) | Test (BlenderKit) | Test (Haven) |
|-------|------------------|--------------------|--------------:|-----------------|-------------------|-------------:|
| Animals | 2,369 | 133 | 1 | 416 | 103 | 1 |
| Appliances | 1,966 | 388 | 5 | 346 | 150 | 2 |
| Architecture | 2,224 | 523 | 7 | 392 | 171 | 3 |
| Decoration | 2,226 | 731 | 0 | 392 | 188 | 0 |
| Electronics | 1,905 | 246 | 6 | 336 | 126 | 3 |
| Furniture | 2,154 | 1,075 | 0 | 380 | 190 | 0 |
| Lighting | 1,565 | 266 | 1 | 278 | 117 | 0 |
| Mechanical | 2,150 | 386 | 18 | 380 | 151 | 8 |
| Nature | 2,782 | 799 | 0 | 492 | 217 | 0 |
| People | 2,554 | 205 | 0 | 452 | 136 | 0 |
| Tools | 1,941 | 645 | 20 | 342 | 152 | 5 |

## Dataset Structure

```
rendr/
├── train/
│ ├── animals/
│ ├── appliances/
│ ├── architecture/
│ ├── decoration/
│ ├── electronics/
│ ├── furniture/
│ ├── lighting/
│ ├── mechanical/
│ ├── nature/
│ ├── people/
│ └── tools/
└── test/
└── [same structure as train]
```

## Normalization Statistics

For standard ImageNet-style normalization:

- **Mean**: `[0.5910, 0.5846, 0.5790]`
- **Std**: `[0.2724, 0.2733, 0.2781]`

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("jneuendorf/rendr")

# Access splits
train_data = dataset['train']
test_data = dataset['test']

# Example: Load with normalization
from torchvision import transforms

transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(
mean=[0.5910, 0.5846, 0.5790],
std=[0.2724, 0.2733, 0.2781]
)
])
```

## Data Sources

- **Rendered Images**: Custom rendered synthetic images
- **BlenderKit**: 3D assets from BlenderKit library
- **Haven**: 3D assets from Poly Haven

## Classes

The dataset includes 11 object categories:
1. Animals
2. Appliances
3. Architecture
4. Decoration
5. Electronics
6. Furniture
7. Lighting
8. Mechanical
9. Nature
10. People
11. Tools

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{rendr,
title={RENDR: A Large-Scale Synthetic 3D Object Dataset},
author={Jim Neuendorf},
year={2025}
}
```

## License

MIT License - Copyright (c) 2025 Jim Neuendorf