Permalink
Browse files

update readme

  • Loading branch information...
hanxiao committed Aug 25, 2017
1 parent 5e60fd2 commit e5295320892c5d4e171d69be400bddeb0ada9b70
Showing with 5 additions and 28 deletions.
  1. +2 −0 .gitignore
  2. +3 −28 README.md
View
@@ -88,3 +88,5 @@ ENV/
.ropeproject
.idea
/data/fashion/holdout-labels-idx1-ubyte.gz
/data/fashion/holdout-images-idx3-ubyte.gz
View
@@ -12,10 +12,7 @@ Here is an example how the data looks like (*each class takes three-columns*):
The original [MNIST dataset](http://yann.lecun.com/exdb/mnist/) contains a lot of handwritten digits. People from AI/ML/Data Science community love this dataset and use it as a benchmark to validate their algorithms. In fact, MNIST is often the first dataset they would try on. *"If it doesn't work on MNIST, it **won't work** at all"*, they said. *"Well, if it does work on MNIST, it may still fail on others."*
We, [Zalando Research](https://research.zalando.com), want to boost the interest of AI inside the company. We would be happy to see our colleagues trying machine learning and pattern recognition techniques on real-world data while spending minimal efforts on preprocessing and formatting. As the Europe's largest online fashion platform, having a data set about fashion articles is definitely more interesting than recognizing digits.
`Fashion-MNIST` is intended to serve as a direct drop-in
replacement for the original MNIST dataset for benchmarking machine learning algorithms as it shares the same image size and the structure of training and testing splits.
`Fashion-MNIST` is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms as it shares the same image size and the structure of training and testing splits.
### To Serious Machine Learning Researchers
@@ -103,9 +100,7 @@ parameters. [Results can be found here.](http://fashion-mnist.s3-website.eu-cent
<img src="doc/img/benchmark.gif" width="100%">
Before submitting a benchmark, please make sure it is not
listed [in this list]((http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/)).
And then please create a new issue, your results will be listed here. Check out the [Contributing](https://github.com/zalandoresearch/fashion-mnist#contributing) section for details.
Before submitting a benchmark, please make sure it is not listed [in this list]((http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/)). And then please create a new issue, your results will be listed here. Check out the [Contributing](https://github.com/zalandoresearch/fashion-mnist#contributing) section for details.
| Classifier | Preprocessing | Test accuracy (mean & std.) | Submitter| Reference|
| --- | --- | --- | --- | --- |
@@ -121,29 +116,9 @@ And then please create a new issue, your results will be listed here. Check out
<img src="doc/img/f04ba662.png" width="50%"><img src="doc/img/4433f0e1.png" width="50%">
## Build Dataset From Scratch
:bomb: It may take hours to rebuild the dataset from scratch. You will need to install the following dependencies:
- `jq`: a json parser
- `pv`: bash pipe viewer
- `parallel`: bash things in parallel
- `ImageMagick`: powerful CLI for image manipulation
- `PIL`: a python image processing package
Make sure you are using Python 3.x.
```bash
cd builder
python build_skuinfo.py
./download_img.sh
python build_dataset.py
```
## Contributing
Thanks for your interest in contributing! There are many ways to contribute to this project. [Get started here!](/CONTRIBUTING.md)
And please check these [open issues](https://github.com/zalandoresearch/fashion-mnist/issues) for specific tasks.
Thanks for your interest in contributing! There are many ways to contribute to this project. [Get started here!](/CONTRIBUTING.md) And please check these [open issues](https://github.com/zalandoresearch/fashion-mnist/issues) for specific tasks.
## Contact
For discussion on the dataset, please use [this Hipchat room.](https://hipchat.zalando.net/chat/room/8093)

0 comments on commit e529532

Please sign in to comment.