Box-based Refinement for Weakly Supervised and Unsupervised Localization Tasks

Eyal Gomel, Tal Shaharbany, Lior Wolf; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 16044-16054

Abstract

It has been established that training a box-based detector network can enhance the localization performance of weakly supervised and unsupervised methods. Moreover, we extend this understanding by demonstrating that these detectors can be utilized to improve the original network, paving the way for further advancements. To accomplish this, we train the detectors on top of the network output instead of the image data and apply suitable loss backpropagation. Our findings reveal a significant improvement in phrase grounding for the "what is where by looking" task, as well as various methods of unsupervised object discovery.

Related Material

[pdf] [supp]
[bibtex]
@InProceedings{Gomel_2023_ICCV, author = {Gomel, Eyal and Shaharbany, Tal and Wolf, Lior}, title = {Box-based Refinement for Weakly Supervised and Unsupervised Localization Tasks}, booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)}, month = {October}, year = {2023}, pages = {16044-16054} }