Variational Feature Pyramid Networks

Panagiotis Dimitrakopoulos, Giorgos Sfikas, Christophoros Nikou
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:5142-5152, 2022.

Abstract

Recent architectures for object detection adopt a Feature Pyramid Network as a backbone for deep feature extraction. Many works focus on the design of pyramid networks which produce richer feature representations. In this work, we opt to learn a dataset-specific architecture for Feature Pyramid Networks. With the proposed method, the network fuses features at multiple scales, it is efficient in terms of parameters and operations, and yields better results across a variety of tasks and datasets. Starting by a complex network, we adopt Variational Inference to prune redundant connections. Our model, integrated with standard detectors, outperforms the state-of-the-art feature fusion networks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-dimitrakopoulos22a, title = {Variational Feature Pyramid Networks}, author = {Dimitrakopoulos, Panagiotis and Sfikas, Giorgos and Nikou, Christophoros}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {5142--5152}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/dimitrakopoulos22a/dimitrakopoulos22a.pdf}, url = {https://proceedings.mlr.press/v162/dimitrakopoulos22a.html}, abstract = {Recent architectures for object detection adopt a Feature Pyramid Network as a backbone for deep feature extraction. Many works focus on the design of pyramid networks which produce richer feature representations. In this work, we opt to learn a dataset-specific architecture for Feature Pyramid Networks. With the proposed method, the network fuses features at multiple scales, it is efficient in terms of parameters and operations, and yields better results across a variety of tasks and datasets. Starting by a complex network, we adopt Variational Inference to prune redundant connections. Our model, integrated with standard detectors, outperforms the state-of-the-art feature fusion networks.} }
Endnote
%0 Conference Paper %T Variational Feature Pyramid Networks %A Panagiotis Dimitrakopoulos %A Giorgos Sfikas %A Christophoros Nikou %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-dimitrakopoulos22a %I PMLR %P 5142--5152 %U https://proceedings.mlr.press/v162/dimitrakopoulos22a.html %V 162 %X Recent architectures for object detection adopt a Feature Pyramid Network as a backbone for deep feature extraction. Many works focus on the design of pyramid networks which produce richer feature representations. In this work, we opt to learn a dataset-specific architecture for Feature Pyramid Networks. With the proposed method, the network fuses features at multiple scales, it is efficient in terms of parameters and operations, and yields better results across a variety of tasks and datasets. Starting by a complex network, we adopt Variational Inference to prune redundant connections. Our model, integrated with standard detectors, outperforms the state-of-the-art feature fusion networks.
APA
Dimitrakopoulos, P., Sfikas, G. & Nikou, C.. (2022). Variational Feature Pyramid Networks. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:5142-5152 Available from https://proceedings.mlr.press/v162/dimitrakopoulos22a.html.

Related Material