Skip to content

Latest commit

 

History

History
149 lines (105 loc) · 8.39 KB

README.md

File metadata and controls

149 lines (105 loc) · 8.39 KB
Raw

Deep Armocromia Dataset visitor badge

University of Macerata, Vision Robotics and Artificial Intelligence (VRAI) Lab

Lorenzo Stacchio, Marina Paolanti, Francesca Spigarelli, Emanuele Frontoni.

[Paper] [Reference]

Figure 1: Deep Armocromia examples.

Deep Armocromia is a novel dataset comprising labeled face images categorized according to Armocromia Flow Theory, with a strict annotation protocol in collaboration with experts.

We conducted experiments to validate the effectiveness of DL models in discriminating among Armocromia classes optimized on Deep Armocromia. Results underscore the challenges inherent to Armocromia classification and highlight opportunities for advancing DL architectures and optimization methodologies.

We release this dataset here for scientific and research purposes only.

Announcements

  • 2024-09 The dataset requests will be accomplished after the paper presentation at ECCV 2024!
  • 2024-08 The paper is accepted at ECCV 2024 Fashion AI Workshop !

Dataset Download

The Deep Armocromia dataset is available in Deep Armocromia Google Drive Folder. You need to fill in the form to get the password for unzipping files. Please take a look at the Data Description below for detailed information about the dataset.

Please note that the labels for the 4 Season labeling were written in Italian, their English translation here follows:

  • Primavera: Spring
  • Estate: Summer
  • Autunno: Autumn
  • Inverno: Winter

Dataset Organization

Each image has a unique entry into the corresponding annotations.csv file which provides the following annotations:

class

  • Description: Represents the primary classification label for the dataset entry.
  • Example: autunno
  • Notes: In this context, it appears to refer to a color season classification.

sub_class

  • Description: Indicates a secondary classification within the primary class, providing more detailed categorization.
  • Example: deep
  • Notes: It specifies a sub-category within the main class, further refining the classification.

partition

  • Description: Specifies the data partition to which the entry belongs, such as training, validation, or test sets.
  • Example: train
  • Notes: This is used to organize the data for different stages of model development and evaluation.

celeba

  • Description: A boolean value indicating whether the entry belongs to the CelebA dataset.
  • Example: True
  • Notes: This is a metadata field used to identify the origin of the image.

path_rgb_original

  • Description: The file path to the original RGB image in the dataset.
  • Example: MERGED_RGB_original/train/autunno/deep/10306.jpg
  • Notes: This points to the unmodified original version of the image.

path_rgb_masked

  • Description: The file path to the RGB image with applied masks.
  • Example: MERGED_masked_RGB_BBOXED/train/autunno/deep/10306.png
  • Notes: This image has masks applied to it, likely for training purposes involving masked regions.

path_mask

  • Description: The file path to the mask image used in the dataset.
  • Example: MERGED_mask_BBOXED/train/autunno/deep/10306.png
  • Notes: This path points to the binary mask image, indicating the regions of interest within the original image.

Please note that all the images amounting to fashion parsing binary masks were extracted with the pre-trained models released in the FACER toolbox, adopting the method reported in the following Figure 2:

Figure 2: Dataset Face Parsing Mask Extraction process.

All the generated mask were analyzed by human experts, that removed the uncorrectly predicted face parsing segmentation masks.

Dataset Statistics

Figure 3 shows the statistics of images and annotations in Deep Armocromia, according to the 4 color seasons and 12 seasonal sub-types (For more precise statistics please refer to the manuscript).

Figure 3: Deep Armocromia Dataset Statistics.

Reference

If you use the DeepArmocromia dataset in your work, please cite it as:

Bibtex

@inproceedings{stacchio2024deep,
  title={Deep Armocromia: A Novel Dataset for Face Seasonal Color Analysis and Classification},
  author={Lorenzo Stacchio and
          Marina Paolanti and
          Francesca Spigarelli and
          Emanuele Frontoni},
  booktitle={European Conference on Computer Vision (ECCV) Workshops},
  pages={xxx--yyy},
  year={2024},
  organization={Springer}
}

APA

Stacchio, L., Paolanti, M., Spigarelli, F., & Frontoni, E. (2024). Deep Armocromia: A novel dataset for face seasonal color analysis and classification. In European Conference on Computer Vision (ECCV) Workshops, (pp. xxx-yyy). Springer.