Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/xu-justin/patchmentation-dataset

This datasets are used to benchmark patch augmentation performance of patchmentation.
https://github.com/xu-justin/patchmentation-dataset

benchmarking dataset-augmentation patchmentation

Last synced: 6 days ago
JSON representation

This datasets are used to benchmark patch augmentation performance of patchmentation.

Host: GitHub
URL: https://github.com/xu-justin/patchmentation-dataset
Owner: Xu-Justin
Created: 2022-12-12T07:08:02.000Z (almost 2 years ago)
Default Branch: main
Last Pushed: 2023-03-07T17:03:29.000Z (over 1 year ago)
Last Synced: 2023-05-19T07:33:53.474Z (over 1 year ago)
Topics: benchmarking, dataset-augmentation, patchmentation
Language: Python
Homepage: https://github.com/Xu-Justin/patchmentation
Size: 3.14 MB
Stars: 1
Watchers: 1
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# Patchmentation Dataset

This datasets are used to benchmark patch augmentation performance of [patchmentation](https://github.com/Xu-Justin/patchmentation).

The benchmarking results can be found at [Xu-Justin/patchmentation-yolov5](https://github.com/Xu-Justin/patchmentation-yolov5#experiment-results-and-comparison).

## Dependency

* Using PIP

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

* Using Docker (recommended)

```bash
docker pull jstnxu/patchmentation:dataset
docker run -it \
-v {cache_folder}:/root/.cache/patchmentation-data \
-v {data_folder}:/workspace/data \
jstnxu/patchmentation:dataset /bin/bash
```

* change `{cache_folder}` to local path to save cache.

* change `{data_folder}` to local path to save generated data.

## Dataset Spesification

* **Training Dataset**

train-pascal-voc-2007

* Number of Images: 2501

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

```bash
python3 dataset.py --version train-pascal-voc-2007 --generate
```

train-pascal-voc-2007-tiny

* Number of Images: 200

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

```bash
python3 dataset.py --version train-pascal-voc-2007-tiny --generate --batch 2
```

train-pascal-voc-2007-v1

* Number of Images: 2,500

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

* Actions

* `filter.FilterWidth(50, Comparator.GreaterEqual)`

* `filter.FilterHeight(50, Comparator.GreaterEqual)`

* `transform.RandomResize(width_range=(50, 150), aspect_ratio=transform.Resize.AUTO_ASPECT_RATIO)`

* Kwargs

* `max_n_patches = 10`

```bash
python3 dataset.py --version train-pascal-voc-2007-v1 --generate --batch 30
```

train-pascal-voc-2007-v2

* Number of images: 2,500

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

* Actions

* `filter.FilterWidth(30, Comparator.GreaterEqual)`

* `filter.FilterHeight(30, Comparator.GreaterEqual)`

* `transform.SoftEdge(13, 20)`

* Kwargs

* `max_n_patches = 20`

* `visibility_threshold = 1.0`

```bash
python3 dataset.py --version train-pascal-voc-2007-v2 --generate --batch 30
```

train-pascal-voc-2007-v3

* Number of images: 2,500

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

* Actions

* Kwargs

* `max_n_patches = 20`

* `visibility_threshold = 0.8`

* `ratio_negative_patch = 5.0`

* `iou_negative_patch = 0.2`

```bash
python3 dataset.py --version train-pascal-voc-2007-v3 --generate --batch 30
```

train-pascal-voc-2007-v4

* Number of images: 2,500

* Number of Classes: 20

* Source: Pascal VOC 2007 - Train

* Actions

* `filter.FilterWidth(30, Comparator.GreaterEqual)`

* `filter.FilterHeight(30, Comparator.GreaterEqual)`

* `transform.SoftEdge(13, 20)`

* Kwargs

* `max_n_patches = 20`

* `visibility_threshold = 0.8`

* `ratio_negative_patch = 5.0`

* `iou_negative_patch = 0.2`

```bash
python3 dataset.py --version train-pascal-voc-2007-v4 --generate --batch 30
```

train-penn-fudan-ped-person

* Number of images: 100

* Number of Classes: 1

* Source: Penn Fudan Ped

```bash
python3 dataset.py --version train-penn-fudan-ped-person --generate --batch 100
```

train-campus

* Number of images: 250

* Number of Classes: 1

* Source: Campus - Garden1, Penn Fudan Ped

* Actions

* `filter.FilterWidth(20, Comparator.GreaterEqual)`

* `filter.FilterHeight(20, Comparator.GreaterEqual)`

* `transform.RandomResize(height_range=(150, 600), aspect_ratio=transform.Resize.AUTO_ASPECT_RATIO)`

* `transform.SoftEdge(5, 10)`

* Kwargs

* `max_n_patches = 30`

* `visibility_threshold = 0.8`

```bash
python3 dataset.py --version train-campus --generate --batch 50
```

* **Validation Dataset**

valid-pascal-voc-2007

* Number of Images: 2,510

* Number of Classes: 20

* Source: Pascal VOC 2007 - Val

```bash
python3 dataset.py --version valid-pascal-voc-2007 --generate
```

valid-penn-fudan-ped-person

* Number of images: 70

* Number of Classes: 1

* Source: Penn Fudan Ped

```bash
python3 dataset.py --version valid-penn-fudan-ped-person --generate
```

valid-campus

* Number of images: 256

* Number of Classes: 1

* Source: Campus - Garden1

```bash
python3 dataset.py --version valid-campus --generate
```

* **Test Dataset**

test-pascal-voc-2007

* Number of Images: 4,952

* Number of Classes: 20

* Source: Pascal VOC 2007 - Test

```bash
python3 dataset.py --version test-pascal-voc-2007 --generate
```

test-campus

* Number of images: 11,538

* Number of Classes: 1

* Source: Campus - Garden1

```bash
python3 dataset.py --version test-campus --generate
```

## Arguments

| Priority* | Arguments | Type | Description |
|:---------:|:--------------:|:-----------------:|------------------------------------------------------------------------------------------------------------------|
| - | `--version` | one or more `str` | Dataset version(s). |
| - | `--overwrite` | `store_true` | Overwrite existing dataset / zip. |
| - | `--batch` | `int` | Number of batch to generate (default=`1`) |
| 1 | `--generate` | `store_true` | Generate the dataset. If `overwrite` is true, it will remove the dataset (if exists) before generating. |
| 2 | `--zip` | `store_true` | Zip the dataset. If `overwrite` is true, it will remove the dataset zip (if exists) before zipping. |
| 3 | `--upload` | `store_true` | Upload the dataset zip. |
| 4 | `--remove-zip` | `store_true` | Remove the dataset zip, if exists. |
| 5 | `--download` | one or more `url` | Download the dataset zip. If `overwrite` is true, it will remove the dataset zip (if exists) before downloading. |
| 6 | `--unzip` | `store_true` | Unzip the dataset zip. If `overwrite` is true, it will remove the dataset (if exists) before unzipping. |
| 7 | `--validate` | `store_true` | Validate the dataset. |
| 8 | `--remove` | `store_true` | Remove the dataset, if exists. |

**Smaller priority number will be executed first*

---

This project was developed as part of thesis project, Computer Science, BINUS University.