File size: 1,463 Bytes
d2be7be
265ee03
 
 
 
 
69aa92e
d2be7be
 
 
 
 
94cfe4a
 
 
d175220
94cfe4a
 
 
 
 
 
 
5b77323
 
 
 
 
 
94cfe4a
 
 
 
 
e2439e6
bf06335
e2439e6
 
 
94cfe4a
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
---
dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype: string
license: cc-by-4.0
task_categories:
- image-classification
pretty_name: Color Multi Fractal DB 1k
---

# Dataset Card for Color Multi Fractal DB 1k

This is a pre-generated 1k classes, 1M images colored-multi-fractal-images dataset based on [Improving Fractal Pre-training](https://catalys1.github.io/fractal-pretraining/) by Connor Anderson et al. and [Multi-Fractal-Dataset](https://github.com/FYGitHub1009/Multi-Fractal-Dataset) by FYSignate1009.

We have changed some fractal parameters so that our ViT pretraining can converge. Modified parameters can be found on this repo.

You can pretrain vision transformers without worrying about dataset licensing for commercial use.

## Dataset Details

### Dataset Sources

- **Repository:** [Improving Fractal Pre-training](https://github.com/catalys1/fractal-pretraining) and [Multi-Fractal-Dataset](https://github.com/FYGitHub1009/Multi-Fractal-Dataset) 
- **Paper:** https://arxiv.org/abs/2110.03091

### Description

- **Curated by:** [ELAN MITSUA Project](https://elanmitsua.com/en/) / Abstract Engine
- **Dataset License:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
- **Source Code License:** [MIT License](https://opensource.org/license/mit)

## How to reproduce this Dataset
This repo contains all scripts that are used to generate fractal images.
```
python generator_no_mixup.py
```