A Review of Deep Learning Models for Computer Vision

Shah, Urmil; Harpale, Aishwarya

doi:10.1109/punecon.2018.8745417

Cited by 9 publications

(8 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our system also differs from our competitors that use contrastive learning (Islam et al, 2021;Feng et al, 2021), that utilizes the similarity and differences in data to improve model training efficiency, to train all of their model parameters. Contrasting our competitors, our classifier backbone consist of pretrained models (Canziani et al, 2016;Shah and Harpale, 2018) that were trained on another task with its head finetuned for paintings classification.…”

Section: Model Agnostic Data Augmentationmentioning

confidence: 99%

“…The classifier, depicted in Figure 5, is made from a pre-trained image classification model like VGG-16 and ResNet-50 (Canziani et al, 2016;Shah and Harpale, 2018) followed by extracting the very first layer and selecting 3 layers between the first and last layers to correspond to features with more spatial information to represent richer features and create a balance between the style and content information's contribution to the classification loss. The spatial attention module takes the re-projected layer for computing attention with the global feature from the bottle neck.…”

Section: Spatial Attention Based Image Classifiermentioning

confidence: 99%

“…We achieve comparable results on the Kaokore dataset with the SOTA accuracy score of 89.04% after 90 epochs using the LOOK method (Feng et al, 2021) as compared to our system with 83.22% after 20 epochs and with a model that requires less training parameters. By changing the proportion of p 1 and p 2 , we can achieve 78.67% precision and 75.3% recall with a ResNet-50 (Shah and Harpale, 2018) backbone. We analyze trends from different proportions of augmentations for the majority and minority classes and check its effectiveness for classifiers with different representation capacities.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Tackling Data Bias in Painting Classification with Style Transfer

Vijendran¹,

Li²,

Shum³

2023

Preprint

View full text Add to dashboard Cite

It is difficult to train classifiers on paintings collections due to model bias from domain gaps and data bias from the uneven distribution of artistic styles. Previous techniques like data distillation, traditional data augmentation and style transfer improve classifier training using task specific training datasets or domain adaptation. We propose a system to handle data bias in small paintings datasets like the Kaokore dataset while simultaneously accounting for domain adaptation in fine-tuning a model trained on real world images. Our system consists of two stages which are style transfer and classification. In the style transfer stage, we generate the stylized training samples per class with uniformly sampled content and style images and train the style transformation network per domain. In the classification stage, we can interpret the effectiveness of the style and content layers at the attention layers when training on the original training dataset and the stylized images. We can tradeoff the model performance and convergence by dynamically varying the proportion of augmented samples in the majority and minority classes. We achieve comparable results to the SOTA with fewer training epochs and a classifier with fewer training parameters.

show abstract

Section: Model Agnostic Data Augmentationmentioning

confidence: 99%

Section: Spatial Attention Based Image Classifiermentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Tackling Data Bias in Painting Classification with Style Transfer

Vijendran¹,

Li²,

Shum³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…In recent years, with the participation of giant technology companies in this competition, its popularity is increasing day by day. ILSVRC competition contributes both to the rapid advancement of the latest technology for computer vision tasks and to the development of general innovations in the architecture of CNN models [4], [11].…”

Section: Deep Learning Modelsmentioning

confidence: 99%

“…Determining and classifying the coordinates of the classes in the images is important at this point. The task of detecting and tracking people and vehicles from the images obtained by drone can be overcome with the deep learning algorithms that have recently been presented [4].…”

Section: Introductionmentioning

confidence: 99%

Derin Öğrenme Tabanlı Transfer Öğrenme Yöntemiyle İnsan ve Araçların Sınıflandırılması

Cengiz

Yılmaz

Kahraman

2021

Düzce Üniversitesi Bilim Ve Teknoloji Dergisi

View full text Add to dashboard Cite

There has been a significant increase in the use of deep learning algorithms in recent years. Convolutional neural network (CNN), one of the deep learning models, is frequently used in applications to distinguish important objects such as humans and vehicles from other objects, especially in image processing. With the development of image processing hardware, the image processing process is significantly reduced. Thanks to these developments, the performance of studies on deep learning is increasing. In this study, a system based on deep learning has been developed to detect and classify objects (human, car and motorcycle / bicycle) from images captured by drones. Two datasets, the image set of Stanford University and the drone image set created at Afyon Kocatepe University (AKÜ), are used to train and test the deep neural network with the transfer learning method. The precision, recall and f1 score values are evaluated according to the process of determining and classifying human, car and motorcycle / bicycle classes using GoogleNet, VggNet and ResNet50 deep learning algorithms. According to this evaluation result, high performance results are obtained with 0.916 precision, 0.895 recall and 0.906 f1 score value in the ResNet50 model.

show abstract