Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/piorosen/implement-mnist
Keras H5 to deploy on "Pure C++"
https://github.com/piorosen/implement-mnist
cpp docker h5 keras mnist tensorflow
Last synced: 21 days ago
JSON representation
Keras H5 to deploy on "Pure C++"
- Host: GitHub
- URL: https://github.com/piorosen/implement-mnist
- Owner: Piorosen
- License: mit
- Created: 2024-06-07T23:15:08.000Z (8 months ago)
- Default Branch: main
- Last Pushed: 2024-06-08T12:34:53.000Z (8 months ago)
- Last Synced: 2024-11-22T00:49:23.462Z (3 months ago)
- Topics: cpp, docker, h5, keras, mnist, tensorflow
- Language: C++
- Homepage:
- Size: 222 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# MNIST for Pure C++ from H5 model on Keras
______ __ __ _______ __ ________ __ __ ________ __ __ ________
/ |/ \ / |/ \ / | / |/ \ / |/ |/ \ / |/ |
$$$$$$/ $$ \ /$$ |$$$$$$$ |$$ | $$$$$$$$/ $$ \ /$$ |$$$$$$$$/ $$ \ $$ |$$$$$$$$/
$$ | $$$ \ /$$$ |$$ |__$$ |$$ | $$ |__ $$$ \ /$$$ |$$ |__ $$$ \$$ | $$ |
$$ | $$$$ /$$$$ |$$ $$/ $$ | $$ | $$$$ /$$$$ |$$ | $$$$ $$ | $$ |
$$ | $$ $$ $$/$$ |$$$$$$$/ $$ | $$$$$/ $$ $$ $$/$$ |$$$$$/ $$ $$ $$ | $$ |
_$$ |_ $$ |$$$/ $$ |$$ | $$ |_____ $$ |_____ $$ |$$$/ $$ |$$ |_____ $$ |$$$$ | $$ |
/ $$ |$$ | $/ $$ |$$ | $$ |$$ |$$ | $/ $$ |$$ |$$ | $$$ | $$ |
$$$$$$/ $$/ $$/ $$/ $$$$$$$$/ $$$$$$$$/ $$/ $$/ $$$$$$$$/ $$/ $$/ $$/
/ \ / |/ \ / |/ | / \ / |
$$ \ /$$ |$$ \ $$ |$$$$$$/ /$$$$$$ |$$$$$$$$/
$$$ \ /$$$ |$$$ \$$ | $$ | $$ \__$$/ $$ |
$$$$ /$$$$ |$$$$ $$ | $$ | $$ \ $$ |
$$ $$ $$/$$ |$$ $$ $$ | $$ | $$$$$$ | $$ |
$$ |$$$/ $$ |$$ |$$$$ | _$$ |_ / \__$$ | $$ |
$$ | $/ $$ |$$ | $$$ |/ $$ |$$ $$/ $$ |
$$/ $$/ $$/ $$/ $$$$$$/ $$$$$$/ $$/
# How to build and Test?
```sh
# if you have docker or other container.
$ git clone https://github.com/Piorosen/implement-mnist.git
$ cd implement-mnist && ./scripts/run.sh --all
```
# How do run it?
#### Docker Build Time: about 40m ~ 60m
```sh
$ git clone https://github.com/Piorosen/implement-mnist.git
# Follow below code, if you want to without building image.
$ docker pull ghcr.io/piorosen/implement-mnist:1.0.0
# following below code, if you want to build by your hand
$ docker build -t ghcr.io/piorosen/implement-mnist:1.0.0 .
$ docker run --rm -it ghcr.io/piorosen/implement-mnist:1.0.0 /bin/ash
$ (container)$ ./mnist # you can see a list of samples in containers.
$ (container)$ ./mnist file 4-0059.txt # need to select file in list.
# Output 4
$ (container)$ cat ./tests/4-0059.txt # if you want to see data.
```
# How do work?
The repository is structured to dynamically generate C++ code to respond to changing weight files.
Therefore, if you try to build C++ without the prerequisites being satisfied, you will get an error in `main.cpp` on line 7 because `#include ‘weights.hpp’` is missing
Therefore, we need to first run `./scripts/model_reproducibility.dockerfile`, a docker file that trains the deep learning model and converts it to C++ code, and import `weight.hpp` separately with `docker cp $container_id:/sources/output/weights.hpp .`.
After that, you can build the `mnist`.
If you are not familiar with `docker`, `Linux`, or `CMakeLists`, you can run `./resources/mnist.cpp`.
The parameter to run is `./mnist file ‘${something file}’`, or you can enter the image directly into the standard input pipeline with `./mnist input`.
## Output
Description|Image
:---:|:---:
h5 keras to cpp|
result of execute|
## AI Model
```
Input (1 x 28 x 28) ->
Average Pooling (1 x 7 x 7) ->
Flatten (1 x 49) ->
Dense (1 x 49 x 30) ->
ReLu (Activate) ->
Dense (1 x 30 x 10) ->
Output (1 x 10)
```
## For Reproducibility
To ensure the reproducibility of the software, I experimented with the random seed value set to 142 in tensorflow. Additionally, when running through docker (same as run.sh --all), it is automatically set to 142. If you reading this does not need reproducibility, please change the argument value in `./scripts/model_reproducibility.dockerfile`.
# [Algorithm Competition](http://ascode.org/problem.php?id=1450)
## Example of Input
input is too long text. [[HERE]](/resources/example_input.txt)
### Input Description
```
1. read first line, test case (< 100)
2. read next line, size 28 x 28 array all of them is float and the range of [0, 1].
3. repeat 2.
```
## Example of Output
`5`
### Output Description
```
Predict number from input data.
```
## Limit in Judge System
```
1. Don't install other library.
2. The length of source code that can be written is limited. (about 30kb)
```