Systematic evaluation of convolution neural network advances on the Imagenet

By Dmytro Mishkin, Nikolay Sergievskiy, Jiri Matas. 2017

265 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

In Evaluation & Benchmarks

Citations in the corpus

ScaleNet: An Unsupervised Representation Learning Method for Limited Information (2023, background): “However, training a small-scale dataset with large batch size such as 1024 and 2048 leads to a decrease in performance, even if tuning learning rate to a heuristic [35].”
Ensemble Adversarial Training: Attacks and Defenses (2017, background): “Mishkin et al. (2017) ﬁnd that input dropout signiﬁcantly degrades accuracy on ImageNet, so we did not include it in our experiments.”
Active Learning for Visual Question Answering: An Empirical Study (2017, background): “Such trends are invariant to choice of image feature [15] and is also observed in ImageNet image classiﬁcation [24].”
Adaptive Estimators Show Information Compression in Deep Neural Networks (2019, background): “, 2011), centered softplus, (Mishkin et al., 2017) and Swish (Ramachandran et al.”
Budgeted Training: Rethinking Deep Neural Network Training Under Resource Constraints (2019, background): “Notably, [44] demonstrates the effectiveness of linear rate decay with CaffeNet on downsized ImageNet.”
Opportunities for neuromorphic computing algorithms and applications (2022, background): “Challenge problems, such as the ImageNet task for deep learning, drove significant advances in that field 111 .”
A Comprehensive Analysis of Deep Regression (2018, unknown): “[35] presents an extensive evaluation of ConvNets on ImageNet for image categorization, including the type of non-linearity, pooling variants, network width, classiﬁer design, image pre-processing, and learning parameters.”
Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks (2018, unknown): “A recent experimental study [26] on CNNs finds that it is sufficient to make qualitative conclusions about optimal network structure that hold for the full-sized (224 × 224 image resolution) ImageNet [31] classification task, by using just 128 × 128 (roughly half the original resolution) input image”
Dropout vs. batch normalization: an empirical study of their impact to deep learning (2020, background): “Although [7] adds batch normalization before the non-linearity, subsequent experiments reported that adding batch normalization after the non-linearity improves accuracy [21].”