Imported Upstream version 1.25.0
[platform/core/ml/nnfw.git] / tools / generate_datafile / tf_dataset_converter / README.md
1 # tf dataset converter
2
3 ## What is tf dataset converter?
4
5 _tf dataset converter_ is a tool which converts tensorflow datasets to datasets for `onert_train`.
6
7 ## Possible datasets
8 - Tensorflow datasets with [ClassLabel feature](https://www.tensorflow.org/datasets/api_docs/python/tfds/features/ClassLabel)
9
10 ## Prerequisite
11 - Python 3.8 (python3.8, python3.8-dev packages)
12 - Python packages required
13
14 ## Usage
15 usage: main.py [-h] [-s] [-d Dataset] [-o Dir] [-p Prefix] [-l N] [-t N]
16
17 Convert a dataset of tensorflow to onert format
18
19 options:
20   -h, --help            show this help message and exit
21   -s, --show-datasets   show dataset list
22   -d Dataset, --dataset-name Dataset
23                         name of dataset to be converted (default: "fashion_mnist")
24   -o Dir, --out-dir Dir
25                         relative path of the files to be created (default: "out")
26   -p Prefix, --prefix-name Prefix
27                         prefix name of the file to be created (default: "")
28   -l N, --train-length N
29                         Number of data for training (default: 1000)
30   -t N, --test-length N
31                         Number of data for training (default: 100)
32
33 ## Example
34 ### Install required packages
35 ```
36 $ python3 -m pip install -r requirements.txt
37 ```
38
39 ### Show dataset list
40 ```
41 $ python3 main.py --show-datasets
42 Dataset list :
43 [abstract_reasoning,
44 accentdb,
45 ...
46 fashion_mnist,
47 ...
48 robotics:mt_opt_sd]
49 ```
50
51 ### Convert dataset to onert format
52 ```
53 $ python3 main.py \
54  --dataset-name fashion_mnist \
55  --prefix-name fashion-mnist \
56  --train-length 2000 \
57  --test-length 200
58 ```
59 ```
60 $ tree out
61 out
62 ├── fashion-mnist.test.input.200.bin
63 ├── fashion-mnist.test.output.200.bin
64 ├── fashion-mnist.train.input.2000.bin
65 └── fashion-mnist.train.output.2000.bin
66 ```