Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/galeone/pascifar
Subset of CIFAR 10/100 datasets PASCAL VOC 2012 compatible.
https://github.com/galeone/pascifar
Last synced: 5 days ago
JSON representation
Subset of CIFAR 10/100 datasets PASCAL VOC 2012 compatible.
- Host: GitHub
- URL: https://github.com/galeone/pascifar
- Owner: galeone
- License: mpl-2.0
- Created: 2016-06-11T08:58:04.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2019-11-03T12:43:25.000Z (about 5 years ago)
- Last Synced: 2024-10-30T06:27:31.989Z (about 2 months ago)
- Language: Python
- Size: 12.7 KB
- Stars: 1
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
PASCIFAR
========
Extract the PASCAL VOC 2012 compatible classes from CIFAR 10 and CIFAR 100 train datasets.
Creates a new dataset called PASCIFAR (PASCAL + CIFAR) that can be used to test classification algorithms trained on the PASCAL VOC 2012 dataset.
There are 4 missing categories from the PASCAL VOC 2012 dataset:
1. cow
2. pottedplant
3. sheep
4. boat
PASCIFAR, thus, covers 16/20 PASCAL VOC 2012 classes.
The `ts.csv` file contains the `file/path.png, label_id` where `label_id` has the value that's the position of the label in the PASCAL VOC dataset sorted list of labels.
Using the following format you can avoid to convert labels between datasets.
# Relations
PASCAL VOC 2012 | CIFAR-10 | CIFAR-100
--- | --- | ---
airplane | airplane | -
bicycle | - | bicycle
bird | bird | -
boat | - | -
bottle | - | food containers (bottle)
bus | - | vehicles 1 (bus)
car | car | -
cat | cat | -
chair | - | household furniture (chair)
cow | - | -
diningtable | - | household furniture (table)
dog | dog | -
horse | horse | -
motorbike | - | vehicles 1 (motorcycle)
person | - | people (macroclass)
pottedplan | - | -
sheep | - | -
sofa | - | houseold furniture (couch)
train | - | vehicles 1 (train)
tvmonitor | - | household electrical devices (television)
# Usage
```python
python builder.sh
```
It downloads the required CIFAR datasets and build PASCIFAR in the current directory.
# Dataset structure and content
PASCIFAR uses the simplest structure: 1 folder for each class.
Every image is in `.png` format and have the fixed size of the CIFAR dataset: `32x32x3`.
Each folder has the name of the corresponding PASCAL VOC 2012 class.
Here's the detailed content:
Class | #
--- | ---
car | 5000
cat | 5000
bird | 5000
sofa | 500
person | 2500
bottle | 500
train | 500
horse | 5000
motorbike | 500
chair | 500
bus | 500
dog | 5000
bicycle | 500
tvmonitor | 500
diningtable | 500
aeroplane | 5000
Total | 42000
# Copyright
CIFAR-10 and CIFAR-100 images has been collected by Alex Krizhevsky: https://www.cs.toronto.edu/~kriz/cifar.html
PASCIFAR `builder.py` is released under Mozilla Public License 2.0.