# nn

**Repository Path**: more_than_code_admin/nn

## Basic Information

- **Project Name**: nn
- **Description**: 基于numpy，cupy与少量的cuda实现简单的卷积网络框架，并复现YOLO-V2目标检测算法。
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 13
- **Forks**: 0
- **Created**: 2021-10-28
- **Last Updated**: 2022-09-28

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Pyconv

#### Introduction
A simple convolution neural network framework based on numpy, cupy and CUDA.


#### Installation

Python3 is required for this framework. 

Run the following commands to install third party dependencies.

```shell
pip install -r requirements.txt
```

#### Demo

1.  **Train a digit classifier on Minist**

    <img src="./markdown/demo1.png" style="zoom: 40%;" />

    Take up the following steps:

    1.  Download minist dataset from http://yann.lecun.com/exdb/mnist/ .

    2.  Unpack minist dataset to `minist/data`.

    3.  Move python scripts:

        ```shell
        mv minist/trian.py .
        mv minist/test.py .
        ```

    4.  Run `train.py` to train classifier.

        ```shell
        python train.py
        ```

    5.  Run `test.py` to test trained classifier.

        ```shell
        python test.py
        ```

2.  **Train a smart player to play flappy bird**

    I **swear** that I have finished this demo, but I have no time to merge that code to this repo.

3.  **Train yolo-v2-tiny detector on VOC dataset**

    Now this toy framework has been used to overfit detector on a small dataset, but hasn't been used to train detector on the whole dataset from scratch yet. (no GPU, no time, and i am lazy ... qwq)

    <img src="./markdown/demo3.png" style="zoom: 40%;" />

    Take up the following steps:
    
    1.  Download VOC dataset:
    
        ```shell
        chmod a+x yolov2/scripts/voc2007.sh
        ./yolov2/scripts/voc2007.sh
        ```
    
    2.  Transform dataset from voc-format to darknet-format:

        ```shell
        python yolov2/scripts/voc2yolo.py --path path/to/voc_root --to ./yolov2/data
        ```
    
    3.  Generate anchors for VOC dataset:

        ```shell
        python yolov2/scripts/anchors.py --files path/to/train.txt
        ```
    
    4.  Move python scripts:

        ```shell
        mv yolov2/overfit.py .
        ```
    
    5.  Run `overfit.py` to overfit detector on 4 images:
    
        ```shell
        python overfit.py
        ```