yzhao062
Could PyOD be implemented in the CV datasets in the future?
I appreciate that pyod now have added some cutting-edge techniques in anomaly detection domain like SO-GAAL, MO-GAAL and Deep SVDD. Could these models also be implemented in the computer vision datasets (like MNIST or CIFAR10) in the future? Besides, the improved version of Deep SVDD——Deep SAD may be added in pyod. Thanks!
Thanks for the note. For the CV data, if you prepare them in the dataloader and have the transformation, some models may directly handle them. Yeah, I am aware of deepSAD--just lack the bandwidth to implement. Feel free to contribute if you want :)
Thanks for the note. For the CV data, if you prepare them in the dataloader and have the transformation, some models may directly handle them. Yeah, I am aware of deepSAD--just lack the bandwidth to implement. Feel free to contribute if you want :)
Thanks for the reply! Does the transformation mean that we should flatten the CV images to the tabular form? (e.g., transform the original 1000x32x32 image to a 1000x1024 tabular data). However, implementing the convolutional structure to the existing pyod models may be a better choice.
Yes. This is only for the existing models...transforming to tabular data is needed. I am not sure when will more dl models, e.g., with conv layers, will be released. If you happen to implement some, feel free to contribute :)
Yes. This is only for the existing models...transforming to tabular data is needed. I am not sure when will more dl models, e.g., with conv layers, will be released. If you happen to implement some, feel free to contribute :)
Thanks. Actually, I have tried some anomaly detection models implemented with convolutional layers, e.g., DeepSAD (with LeNet backbone as in the original paper) and GANomaly (with DCGAN backbone), maybe I would contribute to the pyod next time :)))
In my opinion, pyod would be a better package if some anomaly detection models in various domains (Like in tabular, CV, video or NLP) could be applied~