ikaankeskin/PASCAL_MLX

VOC2012 Image and Annotation Visualization Notebook

Github: https://github.com/ikaankeskin/MLXdatasets/tree/main/ObjectDetection/PASCAL

HuggingFace: https://huggingface.co/datasets/ikaankeskin/PASCAL_MLX

This repository contains a tool that facilitates the download, extraction, and visualization of the VOC2012 dataset, complete with bounding box annotations extracted from associated XML files.

Features

• Automated Dataset Download: Fetches the VOC2012 dataset from Hugging Face's repository in ZIP format.
• ZIP Extraction: Conveniently unzips the downloaded dataset to provide access to images and their annotations.
• Image Visualization: Displays a select set of images from the dataset for preliminary visualization.
• XML Annotation Processing: Reads corresponding XML annotation files for chosen images.
• Bounding Box Overlay: Draws bounding boxes around annotated objects on the images, enhancing visualization.
• Annotation Table Display: Offers a structured view of extracted details from XML annotations in tabular format.

- **Python**: Version 3.x
- **Libraries**: As specified in **`requirements.txt`**, which includes:
    - requests
    - tqdm
    - pandas
    - matplotlib
    - opencv-python

## **Object Filters for Visualizations**

The tool comes equipped with a specific color mapping that governs the visual representation of certain objects when overlaying bounding box annotations on images. The current mapping is coded as:

```python
color_mapping = {'train': (0, 255, 0), 'person': (0, 0, 255)}

This implies:

• 'train' objects are rendered with green bounding boxes (RGB: (0, 255, 0)).
• 'person' objects are visualized with blue bounding boxes (RGB: (0, 0, 255)).

Objects not included in this mapping will not receive bounding boxes during visualization. For incorporating additional object types or altering existing color configurations, users can edit or extend the color_mapping dictionary. For instance, to visualize 'car' objects in red, an entry 'car': (255, 0, 0) can be added.