Home

Awesome

XSR-OSDA

:fire: Implementation for the ''Interpretable Novel Target Discovery Through Open-Set Domain Adaptation (XSR-OSDA)'' work (under review).

XSR-OSDA is an extension work of the "SR-OSDA" paper published in ICCV 2021 [paper][Github].

image

Data Preparation


DatasetDomainRole#Images#Attributes#Classes
DomainNet $\rightarrow$ AwA2AwA <br> Paint <br> RealS / T9,343 / 15,306 <br> 3,441 / 5,760 <br> 5,251 / 10,0478510 / 17
I $\rightarrow$ AwAI / AwAS / T2,970 / 37,3228540 / 50
Domain $\rightarrow$ LADLAD <br> Paint <br> RealS / T13,322 / 19,744 <br> 11,714 / 15,311 <br> 22,395 / 31,06625340 / 56

Dependencies


Training


python main.py

Evaluation


$OS^*$: class-wise average accuracy on the seen categories.

$OS^\diamond$: class-wise average accuracy on the unseen categories correctly classified as "unknown".

$OS$: $\frac{OS^* \times C_{shr} + OS^\diamond}{C_{shr} + 1}$

$OS^{H}$: $\frac{ 2 \times OS^* \times OS^\diamond}{OS^* + OS^\diamond}$

$S$: class-wise average accuracy on shared classes

$U$: class-wise average accuracy on unknown classes

$H = \frac{2 \times S \times U}{ S + U}$

Citation


If you think this work is interesting, please cite:

@InProceedings{Jing_2021_XSROSDA,
author = {Jing, Taotao and Xia, Haifeng and Liu, Hongfu and Ding, Zhengming},
title = {Interpretable Novel Target Discovery Through Open-Set Domain Adaptation},
booktitle = {},
year = {}
}

Contact


If you have any questions about this work, feel free to contact