Do Better ImageNet Models Transfer Better?
Preprint
Abstract
Transfer learning is a cornerstone of computer vision, yet little work has been done to evaluate the relationship between architecture and transfer. An implicit hypothesis in modern computer vision research is that models that perform better on ImageNet necessarily perform better on other vision tasks. However, this hypothesis has never been systematically tested. Here, we compare the performance of 16 classification networks on 12 image classification datasets. We find that, when networks are used as fixed feature extractors or fine-tuned, there is a strong correlation between ImageNet accuracy and transfer accuracy ($r = 0.99$ and $0.96$, respectively). In the former setting, we find that this relationship is very sensitive to the way in which networks are trained on ImageNet; many common forms of regularization slightly improve ImageNet accuracy but yield penultimate layer features that are much worse for transfer learning. Additionally, we find that, on two small fine-grained image classification datasets, pretraining on ImageNet provides minimal benefits, indicating the learned features from ImageNet do not transfer well to fine-grained tasks. Together, our results show that ImageNet architectures generalize well across datasets, but ImageNet features are less general than previously suggested.
Keywords
All Related Versions
This publication has 8 references indexed in Scilit:
- Mask-CNN: Localizing parts and selecting descriptors for fine-grained bird species categorizationPattern Recognition, 2018
- Object-Part Attention Model for Fine-Grained Image ClassificationIEEE Transactions on Image Processing, 2017
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFsIEEE Transactions on Pattern Analysis and Machine Intelligence, 2017
- Coarse-to-Fine Description for Fine-Grained Visual CategorizationIEEE Transactions on Image Processing, 2016
- Factors of Transferability for a Generic ConvNet RepresentationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2015
- cocor: A Comprehensive Solution for the Statistical Comparison of CorrelationsPLOS ONE, 2015
- Speeded-Up Robust Features (SURF)Computer Vision and Image Understanding, 2008
- Testing Dependent Correlations With Nonoverlapping Variables: A Monte Carlo SimulationThe Journal of Experimental Education, 2004