Eyecandies Dataset

A Synthetic Dataset For Anomaly Detection


Read More
Eyecandies Dataset | PAPER Eyecandies Dataset

PAPER

The Eyecandies Dataset for Unsupervised Multimodal Anomaly Detection and Localization

Luca Bonfiglioli*, Marco Toschi*, Davide Silvestri, Nicola Fioraio, Daniele De Gregorio
Eyecan.ai * Equal contribution

We present Eyecandies, a novel synthetic dataset for unsupervised anomaly detection and localization. Photo-realistic images of procedurally generated candies are rendered in a controlled environment under multiple lightning conditions, also providing depth and normal maps in an industrial conveyor scenario. We make available anomaly-free samples for model training and validation, while anomalous instances with precise ground-truth annotations are provided only in the test set. The dataset comprises ten classes of candies, each showing different challenges, such as complex textures, self-occlusions and specularities. Furthermore, we achieve large intra-class variation by randomly drawing key parameters of a procedural rendering pipeline, which enables the creation of an arbitrary number of instances with photo-realistic appearance. Likewise, anomalies are injected into the rendering graph and pixel-wise annotations are automatically generated, overcoming human-biases and possible inconsistencies.

We believe this dataset may encourage the exploration of original approaches to solve the anomaly detection task, e.g. by combining color, depth and normal maps, as they are not provided by most of the existing datasets. Indeed, in order to demonstrate how exploiting additional information may actually lead to higher detection performance, we show the results obtained by training a deep convolutional autoencoder to reconstruct different combinations of inputs.

Cite Us

If you use this dataset in your research, please cite the following paper:

@inproceedings{bonfiglioli2022eyecandies,
    title={The Eyecandies Dataset for Unsupervised Multimodal Anomaly Detection and Localization},
    author={Bonfiglioli, Luca and Toschi, Marco and Silvestri, Davide and Fioraio, Nicola and De Gregorio, Daniele},
    booktitle={Proceedings of the 16th Asian Conference on Computer Vision (ACCV2022},
    note={ACCV},
    year={2022},
}

THE DATASET

Ten Object Classes

Candy CaneChocolate CookieChocolate PralineConfettoGummy Bear
Alt textAlt textAlt textAlt textAlt text
Hazelnut TruffleLicorice SandwichLollipopMarshmallowPeppermint Candy
Alt textAlt textAlt textAlt textAlt text

Multi-Modal

RGBDepthNormals
Alt textAlt textAlt text

Multi-Light

All spotlightsBottom spotRight spotTop spotLeft spotGlobal light box
Alt textAlt textAlt textAlt textAlt textAlt text
Candy CaneChocolate CookieChocolate PralineConfettoGummy Bear
Alt textAlt textAlt textAlt textAlt text
Hazelnut TruffleLicorice SandwichLollipopMarshmallowPeppermint Candy
Alt textAlt textAlt textAlt textAlt text

LEADERBOARD

MethodCan. C.Cho. C.Cho. P.Confet.Gum. B.Haz. T.Lic. S.Lollip.Marsh.Pep. C.Avg.
PADIM (wide resnet50)0.5310.8160.8210.8560.8260.7270.7840.6650.9870.9240.794
PADIM (resnet18)0.5370.7650.7540.7940.7980.6450.7520.6210.9780.8940.754
STFPM (wide resnet50)0.5510.6540.5760.7840.7370.7900.7780.620.840.7490.708
SSIM-AE (RGB)0.5270.8480.7720.7340.590.5080.6930.7600.8510.730.701
DFM (wide resnet50)0.5320.7760.6240.6750.6810.5960.6850.6180.9640.770.692
STFPM (resnet18)0.5270.6280.7660.6660.7280.7270.7380.5720.8930.6310.688
DFM (resnet18)0.5290.7590.5870.6490.6550.6110.6920.5990.9420.7360.676
DFKDE (wide resnet50)0.5390.5770.4820.5480.5410.4920.5240.6020.6580.5910.555
DFKDE (resnet18)0.5370.5890.5170.490.5910.490.5320.5360.6460.5180.545
GANomaly0.4850.5120.5320.5040.5580.4860.4670.5110.4810.5280.507


Submit Your Results

Send us your results on the private test set so we can add your method to our leaderboard!

To do so, send an e-mail to info@eyecan.ai with subject “Eyecandies results submission” and the following info:

  • The name/s of your method/s.
  • A link to a published paper describing it/them.
  • A download link with your results for every proposed method, with predicted heatmaps for every test set sample. We will compute metrics on those heatmaps.

Download the template submission for more info on how to create your own and take a look at the Eyecandies repo for examples and tutorials. The dataset is formatted as a Pipelime underfolder, so take a look at Pipelime as well!

Feel free to ask us any questions!

Luca Bonfiglioli

Luca Bonfiglioli

luca.bonfiglioli@eyecan.ai

Marco Toschi

Marco Toschi

marco.toschi@eyecan.ai

Davide Silvestri

Davide Silvestri

davide.silvestri@eyecan.ai

Nicola Fioraio

Nicola Fioraio

nicola.fioraio@eyecan.ai

Daniele De Gregorio

Daniele De Gregorio

daniele.degregorio@eyecan.ai


You Can find us here