Skip to main content

The Untapped Potential of Off-the-Shelf Convolutional Neural Networks

Publication ,  Conference
Inkawhich, M; Inkawhich, N; Davis, E; Li, H; Chen, Y
Published in: Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022
January 1, 2022

Over recent years, a myriad of novel convolutional network architectures have been developed to advance state-of-the-art performance on challenging recognition tasks. As computational resources improve, a great deal of effort has been placed on efficiently scaling up existing designs and generating new architectures with Neural Architecture Search (NAS) algorithms. While network topology has proven to be a critical factor for model performance, we show that significant gains are being left on the table by keeping topology static at inference-time. Due to challenges such as scale variation, we should not expect static models configured to perform well across a training dataset to be optimally configured to handle all test data. In this work, we expose the exciting potential of inference-time-dynamic models. We show that by allowing just four layers to dynamically change configuration at inference-time, off-the-shelf models like ResNet-50 have an upper bound accuracy of over 95% on ImageNet. This level of performance currently exceeds that of models with over 20x more parameters and significantly more complex training procedures. While this upper bound of performance may be practically difficult to achieve for a real dynamic model, it indicates a significant source of untapped potential for current models.

Duke Scholars

Published In

Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022

DOI

Publication Date

January 1, 2022

Start / End Page

2907 / 2916
 

Citation

APA
Chicago
ICMJE
MLA
NLM
Inkawhich, M., Inkawhich, N., Davis, E., Li, H., & Chen, Y. (2022). The Untapped Potential of Off-the-Shelf Convolutional Neural Networks. In Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022 (pp. 2907–2916). https://doi.org/10.1109/WACV51458.2022.00296
Inkawhich, M., N. Inkawhich, E. Davis, H. Li, and Y. Chen. “The Untapped Potential of Off-the-Shelf Convolutional Neural Networks.” In Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022, 2907–16, 2022. https://doi.org/10.1109/WACV51458.2022.00296.
Inkawhich M, Inkawhich N, Davis E, Li H, Chen Y. The Untapped Potential of Off-the-Shelf Convolutional Neural Networks. In: Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022. 2022. p. 2907–16.
Inkawhich, M., et al. “The Untapped Potential of Off-the-Shelf Convolutional Neural Networks.” Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022, 2022, pp. 2907–16. Scopus, doi:10.1109/WACV51458.2022.00296.
Inkawhich M, Inkawhich N, Davis E, Li H, Chen Y. The Untapped Potential of Off-the-Shelf Convolutional Neural Networks. Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022. 2022. p. 2907–2916.

Published In

Proceedings - 2022 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2022

DOI

Publication Date

January 1, 2022

Start / End Page

2907 / 2916