CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters
- Currently, many theoretical as well as practically relevant questions towards the transferability and robustness of Convolutional Neural Networks (CNNs) remain unsolved. While ongoing research efforts are engaging these problems from various angles, in most computer vision related cases these approaches can be generalized to investigations of the effects of distribution shifts in image data. InCurrently, many theoretical as well as practically relevant questions towards the transferability and robustness of Convolutional Neural Networks (CNNs) remain unsolved. While ongoing research efforts are engaging these problems from various angles, in most computer vision related cases these approaches can be generalized to investigations of the effects of distribution shifts in image data. In this context, we propose to study the shifts in the learned weights of trained CNN models. Here we focus on the properties of the distributions of dominantly used 3×3 convolution filter kernels. We collected and publicly provide a dataset with over 1.4 billion filters from hundreds of trained CNNs, using a wide range of datasets, architectures, and vision tasks. In a first use case of the proposed dataset, we can show highly relevant properties of many publicly available pre-trained models for practical applications: I) We analyze distribution shifts (or the lack thereof) between trained filters along different axes of meta-parameters, like visual category of the dataset, task, architecture, or layer depth. Based on these results, we conclude that model pre-training can succeed on arbitrary datasets if they meet size and variance conditions. II) We show that many pre-trained models contain degenerated filters which make them less robust and less suitable for fine-tuning on target applications. Data & Project website: https://github.com/paulgavrikov/cnn-filter-db.…
Document Type: | Conference Proceeding |
---|---|
Conference Type: | Konferenzartikel |
Zitierlink: | https://opus.hs-offenburg.de/6441 | Bibliografische Angaben |
Title (English): | CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters |
Conference: | 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 18-24 June 2022, New Orleans, LA, USA |
Author: | Paul GavrikovStaff MemberORCiDGND, Janis KeuperStaff MemberORCiDGND |
Year of Publication: | 2022 |
Publisher: | IEEE |
First Page: | 19044 |
Last Page: | 19054 |
Parent Title (English): | Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022) |
ISBN: | 978-1-6654-6946-3 (Elektronisch) |
ISBN: | 978-1-6654-6947-0 (Print on Demand) |
ISSN: | 2575-7075 (Elektronisch) |
ISSN: | 1063-6919 (Print on Demand) |
DOI: | https://doi.org/10.1109/CVPR52688.2022.01848 |
Language: | English | Inhaltliche Informationen |
Institutes: | Fakultät Elektrotechnik, Medizintechnik und Informatik (EMI) (ab 04/2019) |
Forschung / IMLA - Institute for Machine Learning and Analytics | |
Institutes: | Bibliografie |
Tag: | CNN | Formale Angaben |
Relevance: | Konferenzbeitrag: h5-Index > 30 |
Open Access: | Closed |
Licence (German): | Urheberrechtlich geschützt |
ArXiv Id: | http://arxiv.org/abs/2203.15331v2 |