Timur Akhtyamov - Academia.edu (original) (raw)
Uploads
Papers by Timur Akhtyamov
2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Figure 1. From top-left: examples of frames from the SmartPortraits dataset videos that capture h... more Figure 1. From top-left: examples of frames from the SmartPortraits dataset videos that capture human portraits in different natural environments, with varying lightning conditions, using a smartphone and external depth camera on a rig. Bottom-right: recorded trajectory (red-initial time, green-end time) and dense reconstruction obtained by ACMP [88].
INTERNATIONAL CONFERENCE ON INFORMATICS, TECHNOLOGY, AND ENGINEERING 2021 (InCITE 2021): Leveraging Smart Engineering
The article considers two approaches to detecting underwater objects in the image, i.e. classical... more The article considers two approaches to detecting underwater objects in the image, i.e. classical approach and neural network approach. Main advantages and disadvantages of each approach are presented. Various approaches to operation quality were analyzed, including assessment of speed and accuracy, as well as identification of preconditions required to achieve the maximum quality. The article also considers preliminary use of image dehazing methods to improve visibility and contrast. Objects for recognition considered in the article are elements of missions in the Singapore Autonomous Underwater Vehicle Challenge (SAUVC) competitions in Singapore. Nvidia Jetson TX2 singleboard computer is the target platform for the proposed methods, analysis of the method speed was carried out both using the graphics processing unit (GPU) for neural network, and without using it in classical and neural network methods in order to obtain potential speed estimate on simpler platforms without the GPU.
IEEE Access, 2022
Deep convolutional neural networks are highly efficient for computer vision tasks using plenty of... more Deep convolutional neural networks are highly efficient for computer vision tasks using plenty of training data. However, there remains a problem of small training datasets. For addressing this problem the training pipeline which handles rare object types and an overall lack of training data to build wellperforming models that provide stable predictions is required. This article reports on the comprehensive framework XtremeAugment which provides an easy, reliable, and scalable way to collect image datasets and to efficiently label and augment collected data. The presented framework consists of two augmentation techniques that can be used independently and complement each other when applied together. These are Hardware Dataset Augmentation (HDA) and Object-Based Augmentation (OBA). HDA allows the users to collect more data and spend less time on manual data labeling. OBA significantly increases the training data variability and remains the distribution of the augmented images being close to the original dataset. We assess the proposed approach for the apple spoil segmentation scenario. Our results demonstrate a substantial increase in the model accuracy reaching 0.91 F1-score and outperforming the baseline model for up to 0.62 F1-score for a few-shot learning case in the wild data. The highest benefit of applying XtremeAugment is achieved for the cases where we collect images in the controlled indoor environment, but have to use the model in the wild.
2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Figure 1. From top-left: examples of frames from the SmartPortraits dataset videos that capture h... more Figure 1. From top-left: examples of frames from the SmartPortraits dataset videos that capture human portraits in different natural environments, with varying lightning conditions, using a smartphone and external depth camera on a rig. Bottom-right: recorded trajectory (red-initial time, green-end time) and dense reconstruction obtained by ACMP [88].
INTERNATIONAL CONFERENCE ON INFORMATICS, TECHNOLOGY, AND ENGINEERING 2021 (InCITE 2021): Leveraging Smart Engineering
The article considers two approaches to detecting underwater objects in the image, i.e. classical... more The article considers two approaches to detecting underwater objects in the image, i.e. classical approach and neural network approach. Main advantages and disadvantages of each approach are presented. Various approaches to operation quality were analyzed, including assessment of speed and accuracy, as well as identification of preconditions required to achieve the maximum quality. The article also considers preliminary use of image dehazing methods to improve visibility and contrast. Objects for recognition considered in the article are elements of missions in the Singapore Autonomous Underwater Vehicle Challenge (SAUVC) competitions in Singapore. Nvidia Jetson TX2 singleboard computer is the target platform for the proposed methods, analysis of the method speed was carried out both using the graphics processing unit (GPU) for neural network, and without using it in classical and neural network methods in order to obtain potential speed estimate on simpler platforms without the GPU.
IEEE Access, 2022
Deep convolutional neural networks are highly efficient for computer vision tasks using plenty of... more Deep convolutional neural networks are highly efficient for computer vision tasks using plenty of training data. However, there remains a problem of small training datasets. For addressing this problem the training pipeline which handles rare object types and an overall lack of training data to build wellperforming models that provide stable predictions is required. This article reports on the comprehensive framework XtremeAugment which provides an easy, reliable, and scalable way to collect image datasets and to efficiently label and augment collected data. The presented framework consists of two augmentation techniques that can be used independently and complement each other when applied together. These are Hardware Dataset Augmentation (HDA) and Object-Based Augmentation (OBA). HDA allows the users to collect more data and spend less time on manual data labeling. OBA significantly increases the training data variability and remains the distribution of the augmented images being close to the original dataset. We assess the proposed approach for the apple spoil segmentation scenario. Our results demonstrate a substantial increase in the model accuracy reaching 0.91 F1-score and outperforming the baseline model for up to 0.62 F1-score for a few-shot learning case in the wild data. The highest benefit of applying XtremeAugment is achieved for the cases where we collect images in the controlled indoor environment, but have to use the model in the wild.