75 lines
1.6 KiB
Markdown
75 lines
1.6 KiB
Markdown
# 数据说明
|
||
|
||
---
|
||
|
||
## 1.简介
|
||
本文档介绍ImageNet1k和flowers102数据准备过程。
|
||
|
||
## 2.数据集准备
|
||
|
||
数据集 | 训练集大小 | 测试集大小 | 类别数 | 备注|
|
||
:------:|:---------------:|:---------------------:|:-----------:|:-----------:
|
||
[flowers102](https://www.robots.ox.ac.uk/~vgg/data/flowers/102/)|1k | 6k | 102 |
|
||
[ImageNet1k](http://www.image-net.org/challenges/LSVRC/2012/)|1.2M| 50k | 1000 |
|
||
|
||
* 数据格式
|
||
按照如下结构组织数据，其中train_list.txt 和val_list.txt的格式形如
|
||
|
||
```shell
|
||
# 每一行采用"空格"分隔图像路径与标注
|
||
|
||
ILSVRC2012_val_00000001.JPEG 65
|
||
...
|
||
|
||
```
|
||
### ImageNet1k
|
||
从官方下载数据后，按如下组织数据
|
||
|
||
```bash
|
||
PaddleClas/dataset/imagenet/
|
||
|_ train/
|
||
| |_ n01440764
|
||
| | |_ n01440764_10026.JPEG
|
||
| | |_ ...
|
||
| |_ ...
|
||
| |
|
||
| |_ n15075141
|
||
| |_ ...
|
||
| |_ n15075141_9993.JPEG
|
||
|_ val/
|
||
| |_ ILSVRC2012_val_00000001.JPEG
|
||
| |_ ...
|
||
| |_ ILSVRC2012_val_00050000.JPEG
|
||
|_ train_list.txt
|
||
|_ val_list.txt
|
||
```
|
||
### Flowers102
|
||
从[VGG官方网站](https://www.robots.ox.ac.uk/~vgg/data/flowers/102/)下载后的数据，解压后包括
|
||
|
||
```shell
|
||
jpg/
|
||
setid.mat
|
||
imagelabels.mat
|
||
```
|
||
|
||
将以上文件放置在PaddleClas/dataset/flowers102/下
|
||
|
||
通过运行generate_flowers102_list.py生成train_list.txt和val_list.txt
|
||
|
||
```bash
|
||
python generate_flowers102_list.py jpg train > train_list.txt
|
||
python generate_flowers102_list.py jpg valid > val_list.txt
|
||
|
||
```
|
||
按照如下结构组织数据：
|
||
|
||
```bash
|
||
PaddleClas/dataset/flowers102/
|
||
|_ jpg/
|
||
| |_ image_03601.jpg
|
||
| |_ ...
|
||
| |_ image_02355.jpg
|
||
|_ train_list.txt
|
||
|_ val_list.txt
|
||
```
|