Filters








12,434 Hits in 5.8 sec

Rethinking ImageNet Pre-training [article]

Kaiming He, Ross Girshick, Piotr Dollár
2018 arXiv   pre-print
The results are no worse than their ImageNet pre-training counterparts even when using the hyper-parameters of the baseline system (Mask R-CNN) that were optimized for fine-tuning pre-trained models, with  ...  To push the envelope we demonstrate 50.9 AP on COCO object detection without using any external data---a result on par with the top COCO 2017 competition results that used ImageNet pre-training.  ...  It can be sufficient to directly train on the target data if its dataset scale is large enough.  ... 
arXiv:1811.08883v1 fatcat:qv7hctiy3ndopdirjhsbdbcore

Greedy Layerwise Learning Can Scale to ImageNet [article]

Eugene Belilovsky, Michael Eickenberg, Edouard Oyallon
2019 arXiv   pre-print
To our knowledge, this is the first competitive alternative to end-to-end training of CNNs that can scale to ImageNet.  ...  Using a simple set of ideas for architecture and training we find that solving sequential 1-hidden-layer auxiliary problems lead to a CNN that exceeds AlexNet performance on ImageNet.  ...  Supervised end-to-end learning is the standard approach to neural network optimization. However it has potential issues that can be valuable to consider.  ... 
arXiv:1812.11446v3 fatcat:sc7yz4pwsrhzfkxdcp64ijpyji

XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks [article]

Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, Ali Farhadi
2016 arXiv   pre-print
Our binary networks are simple, accurate, efficient, and work on challenging visual tasks. We evaluate our approach on the ImageNet classification task.  ...  We compare our method with recent network binarization methods, BinaryConnect and BinaryNets, and outperform these methods by large margins on ImageNet, more than 16% in top-1 accuracy.  ...  XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks  ... 
arXiv:1603.05279v4 fatcat:dsfl3pwi55fxzcicde65yh4s3y

Do ImageNet Classifiers Generalize to ImageNet? [article]

Benjamin Recht, Rebecca Roelofs, Ludwig Schmidt, Vaishaal Shankar
2019 arXiv   pre-print
We evaluate a broad range of models and find accuracy drops of 3% - 15% on CIFAR-10 and 11% - 14% on ImageNet.  ...  However, accuracy gains on the original test sets translate to larger gains on the new test sets.  ...  while working on this paper.  ... 
arXiv:1902.10811v2 fatcat:d32xzqv56fbxfh26omlfy6g4bi

Systematic evaluation of convolution neural network advances on the Imagenet

Dmytro Mishkin, Nikolay Sergievskiy, Jiri Matas
2017 Computer Vision and Image Understanding  
The paper systematically studies the impact of a range of recent advances in CNN architectures and learning methods on the object categorization (ILSVRC) problem.  ...  We show that the use of 128x128 pixel images is sufficient to make qualitative conclusions about optimal network structure that hold for the full size Caffe and VGG nets.  ...  There has not been much research on the optimal colorspace or pre-processing techniques for CNN.  ... 
doi:10.1016/j.cviu.2017.05.007 fatcat:4u2un2jkp5aapc5wfoxihjxtey

Self-EMD: Self-Supervised Object Detection without ImageNet [article]

Songtao Liu, Zeming Li, Jian Sun
2021 arXiv   pre-print
Our Faster R-CNN (ResNet50-FPN) baseline achieves 39.8% mAP on COCO, which is on par with the state of the art self-supervised methods pre-trained on ImageNet.  ...  More importantly, it can be further improved to 40.4% mAP with more unlabeled images, showing its great potential for leveraging more easily obtained unlabeled data. Code will be made available.  ...  For Mask R-CNN with FPN, the AP increases from 38.5% to 39.3 % (+0.8%), for Mask R-CNN with C4 architecture, the AP increases from 37.9% to 38.5% (+0.6%).  ... 
arXiv:2011.13677v3 fatcat:2dv6eda335gltpnlz2cgv6bkpu

Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? [article]

Kensho Hara, Hirokatsu Kataoka, Yutaka Satoh
2018 arXiv   pre-print
We examine the architectures of various 3D CNNs from relatively shallow to very deep ones on current video datasets.  ...  (ii) The Kinetics dataset has sufficient data for training of deep 3D CNNs, and enables training of up to 152 ResNets layers, interestingly similar to 2D ResNets on ImageNet.  ...  datasets are obviously too small to be used for optimizing CNN representations from scratch.  ... 
arXiv:1711.09577v2 fatcat:vdo7vkhspjfkbe2w5ojho5zj4e

Beyond ImageNet: Deep Learning in Industrial Practice [chapter]

Thilo Stadelmann, Vasily Tolkachev, Beate Sick, Jan Stampfli, Oliver Dürr
2019 Applied Data Science  
We will first present a classical application of CNNs on image-like data, in particular, phenotype classification of cells based on their morphology, and then extend the task to clustering voices based  ...  We will focus on convolutional neural networks (CNNs), which have since the seminal work of Krizhevsky et al. (2012) revolutionized image classification and even started surpassing human performance on  ...  The architecture of the CNN is inspired by the second-best entry of the 2014 ImageNet competition (Simonyan & Zisserman, 2015) .  ... 
doi:10.1007/978-3-030-11821-1_12 fatcat:5uo3qf2uw5e2dlzkjvtapynbhu

Solving ImageNet: a Unified Scheme for Training any Backbone to Top Results [article]

Tal Ridnik, Hussam Lawen, Emanuel Ben-Baruch, Asaf Noy
2022 arXiv   pre-print
We test USI on a wide variety of architectures, including CNNs, Transformers, Mobile-oriented and MLP-only. On all models tested, USI outperforms previous state-of-the-art results.  ...  Hence, we are able to transform training on ImageNet from an expert-oriented task to an automatic seamless routine.  ...  Hence, the optimization process on ImageNet is more sensitive to different hyper-parameters, and the architecture used.  ... 
arXiv:2204.03475v2 fatcat:lorfkniriras5pawocfcttxowu

Diverse Imagenet Models Transfer Better [article]

Niv Nayman, Avram Golbert, Asaf Noy, Tan Ping, Lihi Zelnik-Manor
2022 arXiv   pre-print
A commonly accepted hypothesis is that models with higher accuracy on Imagenet perform better on other downstream tasks, leading to much research dedicated to optimizing Imagenet accuracy.  ...  We demonstrate our results on several architectures and multiple downstream tasks, including both single-label and multi-label classification.  ...  A common practice is to pre-train a model on a large-scale supervised dataset such as ImageNet [68] and fine-tune it on the downstream (target) dataset that is typically of a smaller scale.  ... 
arXiv:2204.09134v1 fatcat:5eq6c327rzhtjpb7iyimrtqrrq

Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet [article]

Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi, Zihang Jiang, Francis EH Tay, Jiashi Feng, Shuicheng Yan
2021 arXiv   pre-print
However, ViT achieves inferior performance to CNNs when trained from scratch on a midsize dataset like ImageNet.  ...  It also outperforms ResNets and achieves comparable performance with MobileNets by directly training on ImageNet.  ...  Though ViT proves the full-transformer architecture is promising for vision tasks, its performance is still inferior to that of similar-sized CNN counterparts (e.g.  ... 
arXiv:2101.11986v3 fatcat:fupiobqjdzfkno3uclf2wu6fbq

XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks [chapter]

Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, Ali Farhadi
2016 Lecture Notes in Computer Science  
Our binary networks are simple, accurate, efficient, and work on challenging visual tasks. We evaluate our approach on the ImageNet classification task.  ...  We compare our method with recent network binarization methods, BinaryConnect and BinaryNets, and outperform these methods by large margins on ImageNet, more than 16% in top-1 accuracy.  ...  This architecture was the first CNN architecture that showed to be successful on ImageNet classification task. This network has 61M parameters.  ... 
doi:10.1007/978-3-319-46493-0_32 fatcat:n757ualbzzc5pof4hcv2e3tema

Resizing Tiny Imagenet: An Iterative Approach Towards Image Classification

Praveen Kumar, Student, Department of Computer Science & Engineering, MVJ College of Engineering, Whitefield, Bangalore, India, Rishabh Singh, Nilesh Kumar Singh, Hitesh Agarwal, Student, Department of Computer Science & Engineering, MVJ College of Engineering, Whitefield, Bangalore, India, Student, Department of Computer Science & Engineering, MVJ College of Engineering, Whitefield, Bangalore, India, Student, Department of Computer Science & Engineering, MVJ College of Engineering, Whitefield, Bangalore, India
2020 International Journal of Innovative Research in Computer Science & Technology  
It is generally considered one of the harder datasets in the domain of image classification.  ...  Stanford's Tiny ImageNet dataset has been around for a while and neural networks have struggled to classify them.  ...  Most importantly, it uses a backpropagation algorithm to take the updated weights back to the head of the model. II. DATASET Tiny ImageNet is a subset of ImageNet.  ... 
doi:10.21276/ijircst.2020.8.6.8 fatcat:fa3fxf3jznh5jpazg3q667gceu

Deep Learning Earth Observation Classification Using ImageNet Pretrained Networks

Dimitrios Marmanis, Mihai Datcu, Thomas Esch, Uwe Stilla
2016 IEEE Geoscience and Remote Sensing Letters  
In this letter, we propose a novel method by considering a pretrained CNN designed for tackling an entirely different classification problem, namely, the ImageNet challenge, and exploit it to extract an  ...  Deep learning methods such as convolutional neural networks (CNNs) can deliver highly accurate classification results when provided with large enough data sets and respective labels.  ...  The importance of the trainable CNN stage is essential due to its high discriminant potential, the system is able to learn complex relations among highly abstract representations and correctly separate  ... 
doi:10.1109/lgrs.2015.2499239 fatcat:xcwx2qiiqrfcdiolmga5q62jta

Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective [article]

Wuyang Chen, Xinyu Gong, Zhangyang Wang
2021 arXiv   pre-print
In NAS-Bench-201 and DARTS search spaces, TE-NAS completes high-quality search but only costs 0.5 and 4 GPU hours with one 1080Ti on CIFAR-10 and ImageNet, respectively.  ...  Neural Architecture Search (NAS) has been explosively studied to automate the discovery of top-performer neural networks.  ...  A one-shot super network can share its parameters to sampled sub-networks and accelerate the architecture evaluations, but it is very heavy and hard to optimize and suffers from a poor correlation between  ... 
arXiv:2102.11535v4 fatcat:pnid7ogihncrvh26roc5cijdcu
« Previous Showing results 1 — 15 out of 12,434 results