graviti
Products
Resources
About us
RAF-DB
Classification
Face
|...
License: Unknown

Overview

Real-world Affective Faces Database (RAF-DB) is a large-scale facial expression database
with around 30K great-diverse facial images downloaded from the Internet. Based on the
crowdsourcing annotation, each image has been independently labeled by about 40 annotators.
Images in this database are of great variability in subjects' age, gender and ethnicity, head
poses, lighting conditions, occlusions, (e.g. glasses, facial hair or self-occlusion), post-processing
operations (e.g. various filters and special effects), etc. RAF-DB has large diversities, large
quantities, and rich annotations, including:

  • 29672 number of real-world images,
  • a 7-dimensional expression distribution vector for each image,
  • two different subsets: single-label subset,
    including 7 classes of basic emotions; two-tab subset, including 12 classes of
    compound emotions,
  • 5 accurate landmark locations, 37 automatic landmark locations,
    bounding box, race, age range and gender attributes annotations per image,
  • baseline classifier outputs for basic emotions and compound emotions.

To be able to objectively measure the performance for the followers' entries, the database
has been split into a training set and a test set where the size of training set is five times
larger than test set, and expressions in both sets have a near-identical distribution.

Sample Images

img

Contect preview

  • Single-label Subset (Basic emotions)
    Single-label Subset

  • Two-tab Subset (Compound emotions)
    Two-tab Subset

For more details of the dataset, please refer to the paper Reliable Crowdsourcing and
DeepLocality-Preserving Learning for Expression Recognition in the Wild

here".
* Please note that the RAF database is partially public. And the other 10k images are neither
basic nor compound emotions which will be released afterwards.

Data Collection

At the very beginning, the images’URLs collected from Flickr were fed into an automatic open-source
downloader to download images in batches. Considering that the results returned by Flickr’s image
search API were in well-structured XML format, from which the URLs can be easily parsed, we then
used a set of keywords (for example: smile, giggle, cry, rage, scared, frightened, terrified,
shocked, astonished, disgust, expressionless) to pick out images that were
related with the six basic emotions plus the neutral emotion. At last, a total of 29672 real-world
facial images are presented in our database. Figure 2 shows the pipeline of data collection

Figure 2. Overview of construction and annotation of RAF-DB.

Overview of construction and annotation of RAF-DB.

Data Annotation

Annotating nearly 30000 images of expression is an extremely difficult and time-consuming
task. Considering the compounded property of real-world expressions, multiple views of
images’ expression state should be collected from different labelers. We therefore
employed 315 annotators (students and staffs from universities) who have been instructed
with one-hour tutorial of psychological knowledge on emotion for an online facial
expression annotation assignment, where they were asked to classify the image into the
most apparent one from seven classes. We developed a website for RAF-DB annotation,
which shows each image with exclusive attribute options. Images were randomly and equally
assigned to each labeler, ensuring that there were no direct correlation among the
images labeled by one person. And each image was assured to be labeled by about 40 independent
labelers. After that, a multi-label annotation result is obtained for each image,
i.e., a seven dimensional vector that each dimension indicates the votes of relevant emotion.

Citation

Please use the following citation when referencing the dataset:

@inproceedings{li2017reliable,
  title={Reliable Crowdsourcing and Deep Locality-Preserving Learning for Expression Recognition
in the Wild},
  author={Li, Shan and Deng, Weihong and Du, JunPing},
  booktitle={2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={2584--2593},
  year={2017},
  organization={IEEE}
}
@article{li2019reliable,
  title={Reliable Crowdsourcing and Deep Locality-Preserving Learning for Unconstrained
Facial Expression Recognition},
  author={Li, Shan and Deng, Weihong},
  journal={IEEE Transactions on Image Processing},
  volume={28},
  number={1},
  pages={356--370},
  year={2019},
  publisher={IEEE}
}
Data Summary
Type
Image,
Amount
29.672K
Size
1875.55GB
Provided by
Weihong Deng
Pattern Recognition and Intelligent System Laboratory (PRIS Lab), Beijing University of Posts and Telecommunications.
| Amount 29.672K | Size 1875.55GB
RAF-DB
Classification
Face
License: Unknown

Overview

Real-world Affective Faces Database (RAF-DB) is a large-scale facial expression database
with around 30K great-diverse facial images downloaded from the Internet. Based on the
crowdsourcing annotation, each image has been independently labeled by about 40 annotators.
Images in this database are of great variability in subjects' age, gender and ethnicity, head
poses, lighting conditions, occlusions, (e.g. glasses, facial hair or self-occlusion), post-processing
operations (e.g. various filters and special effects), etc. RAF-DB has large diversities, large
quantities, and rich annotations, including:

  • 29672 number of real-world images,
  • a 7-dimensional expression distribution vector for each image,
  • two different subsets: single-label subset,
    including 7 classes of basic emotions; two-tab subset, including 12 classes of
    compound emotions,
  • 5 accurate landmark locations, 37 automatic landmark locations,
    bounding box, race, age range and gender attributes annotations per image,
  • baseline classifier outputs for basic emotions and compound emotions.

To be able to objectively measure the performance for the followers' entries, the database
has been split into a training set and a test set where the size of training set is five times
larger than test set, and expressions in both sets have a near-identical distribution.

Sample Images

img

Contect preview

  • Single-label Subset (Basic emotions)
    Single-label Subset

  • Two-tab Subset (Compound emotions)
    Two-tab Subset

For more details of the dataset, please refer to the paper Reliable Crowdsourcing and
DeepLocality-Preserving Learning for Expression Recognition in the Wild

here".
* Please note that the RAF database is partially public. And the other 10k images are neither
basic nor compound emotions which will be released afterwards.

Data Collection

At the very beginning, the images’URLs collected from Flickr were fed into an automatic open-source
downloader to download images in batches. Considering that the results returned by Flickr’s image
search API were in well-structured XML format, from which the URLs can be easily parsed, we then
used a set of keywords (for example: smile, giggle, cry, rage, scared, frightened, terrified,
shocked, astonished, disgust, expressionless) to pick out images that were
related with the six basic emotions plus the neutral emotion. At last, a total of 29672 real-world
facial images are presented in our database. Figure 2 shows the pipeline of data collection

Figure 2. Overview of construction and annotation of RAF-DB.

Overview of construction and annotation of RAF-DB.

Data Annotation

Annotating nearly 30000 images of expression is an extremely difficult and time-consuming
task. Considering the compounded property of real-world expressions, multiple views of
images’ expression state should be collected from different labelers. We therefore
employed 315 annotators (students and staffs from universities) who have been instructed
with one-hour tutorial of psychological knowledge on emotion for an online facial
expression annotation assignment, where they were asked to classify the image into the
most apparent one from seven classes. We developed a website for RAF-DB annotation,
which shows each image with exclusive attribute options. Images were randomly and equally
assigned to each labeler, ensuring that there were no direct correlation among the
images labeled by one person. And each image was assured to be labeled by about 40 independent
labelers. After that, a multi-label annotation result is obtained for each image,
i.e., a seven dimensional vector that each dimension indicates the votes of relevant emotion.

Citation

Please use the following citation when referencing the dataset:

@inproceedings{li2017reliable,
  title={Reliable Crowdsourcing and Deep Locality-Preserving Learning for Expression Recognition
in the Wild},
  author={Li, Shan and Deng, Weihong and Du, JunPing},
  booktitle={2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={2584--2593},
  year={2017},
  organization={IEEE}
}
@article{li2019reliable,
  title={Reliable Crowdsourcing and Deep Locality-Preserving Learning for Unconstrained
Facial Expression Recognition},
  author={Li, Shan and Deng, Weihong},
  journal={IEEE Transactions on Image Processing},
  volume={28},
  number={1},
  pages={356--370},
  year={2019},
  publisher={IEEE}
}
0
Start building your AI now
graviti
wechat-QR
Long pressing the QR code to follow wechat official account

Copyright@Graviti