inlier vs outlier

[jyang68sh]

Hi I am bit confused with the definition of inlier and outlier classes described in your paper.

e.g. For cityscapes, what are the inlier classes and outlier classes?

Thanks

[yyliu01]

Thanks for your interest.

The inlier indicates the original 19 classes, while the outlier denotes the anomalous objects.
In the Cityscapes dataset, there is no outlier (in case we ignore the 255 class). We follow the meta-ood to implement the outlier exposure (OE) to introduce the anomalous objects via extra training data (coco). There is a visualisation in Fig.2, Page 6 of our paper, which grabs the laptop from its original image to the driving scenes.

Regards,
Yuyuan

[yyliu01]

Closing the issue, but feel free to reopen.

[jyang68sh]

@

Thanks for your interest.

The inlier indicates the original 19 classes, while the outlier denotes the anomalous objects. In the Cityscapes dataset, there is no outlier (in case we ignore the 255 class). We follow the meta-ood to implement the outlier exposure (OE) to introduce the anomalous objects via extra training data (coco). There is a visualisation in Fig.2, Page 6 of our paper, which grabs the laptop from its original image to the driving scenes.

Regards, Yuyuan

Hi. Thanks for your answer. I am still reading about your abstention learning module. Is it a kind of active learning?

[yyliu01]

@jyang68sh This work is based on the gambler loss (which belongs to abstention learning), and I don't think it is related to active learning. I recommend reading it (https://arxiv.org/abs/1907.00208) for a better understanding of the background.