The experiments in this study used a system comprising an NVIDIA TESLA P40 graphics processing unit (NVIDIA, Santa Clara, CA, USA), an Intel Xeon E5-2630 v4 CPU (Intel, Santa Clara, CA, USA), 32 GB of RAM, and the Ubuntu 20.04.6 LTS operating system. The programming language used for the experiment was Python (version 3.7.16). The libraries used for preprocessing and training the deep learning models were TensorFlow software (version. 2.6.0), Keras (version. 2.6.0) for various functions supporting deep learning model design and training, unified device architecture (version. 11.2.0) for massive computational processing as a graphics processing device development tool, open computer vision (version. 4.6.0.66) providing various functions for image processing, and matplotlib (version. 3.5.2) for data visualization.
Tesla p40
Manufactured by NVIDIA
Sourced in United States
The Tesla P40 is a high-performance graphics processing unit (GPU) developed by NVIDIA for data center and enterprise applications. The Tesla P40 is designed to accelerate deep learning and machine learning workloads, providing powerful computational capabilities for tasks such as image recognition, natural language processing, and speech recognition.
Automatically generated - may contain errors
Lab products found in correlation
4 protocols using tesla p40
1
Deep Learning Model Development Pipeline
2
Deep Learning on NVIDIA Tesla P40
Check if the same lab product or an alternative is used in the 5 most similar protocols
+ Expand
We performed our experiments on a customized server with 1× Nvidia Tesla P40 graphics processing unit (GPU). Our algorithms were developed with Python 3.6 and TensorFlow 1.12 on a Linux platform.
3
Lung Fissure Segmentation Framework
Check if the same lab product or an alternative is used in the 5 most similar protocols
+ Expand
The IntegrityNet architecture is implemented with the open-source framework Keras25 . The network was trained using NVIDIA GPU card Tesla P40 with 24 GB RAM. Adam optimization26 was used for training with a static learning rate of 0.0002. Tversky loss27 (link) was used with α = 0.05 in order to handle the large class imbalance between background and the incomplete and complete fissure classes that lie along the thin fissure surface (i.e., there are many more voxels labeled background in the output image than voxels assigned to intact or incomplete fissure classes).
During training, random cropping to fixed input size of (128, 128, 64) for each image was used to diversify the data seen in training for each epoch. The effect of this method is to increase the amount of data within the training set without the need for more subject images. A validation set was also used to identify the epoch that produced the best results for a set that was held out from training and testing. The train, test, and validation proportions used were 0.75, 0.15, and 0.10 respectively.
During training, random cropping to fixed input size of (128, 128, 64) for each image was used to diversify the data seen in training for each epoch. The effect of this method is to increase the amount of data within the training set without the need for more subject images. A validation set was also used to identify the epoch that produced the best results for a set that was held out from training and testing. The train, test, and validation proportions used were 0.75, 0.15, and 0.10 respectively.
4
iRadonMap Optimization for Reconstruction
Check if the same lab product or an alternative is used in the 5 most similar protocols
+ Expand
To obtain promising reconstruction performance, the iRadonMap is overall optimized by minimizing the mean square error (MSE), which is defined as follows:
Here, x is the final output of the iRadonMap and x ref is the reference image. N is the number of image pairs used for training. Θ represents the learnable parameters in the iRadonMap. This minimization problem can be solved with various off-the-shelf algorithms, and in this work the RMSProp algorithm (see http://www.cs.toronto.edu/∼tijmen/csc321/slides/lecture slides lec6.pdf ) is adopted. The corresponding minibatch size, learning rate, momentum, and weight decay are set to 2, 0.00002, 0.9, and 0.0, respectively. The iRadonMap is implemented on PyTorch deep learning framework. 10 The iRadonMap is trained for one week using two NVIDIA Tesla P40 graphics processing units (GPUs) with 24 GB memory capacity each.
Here, x is the final output of the iRadonMap and x ref is the reference image. N is the number of image pairs used for training. Θ represents the learnable parameters in the iRadonMap. This minimization problem can be solved with various off-the-shelf algorithms, and in this work the RMSProp algorithm (see http://www.cs.toronto.edu/∼tijmen/csc321/slides/lecture slides lec6.pdf ) is adopted. The corresponding minibatch size, learning rate, momentum, and weight decay are set to 2, 0.00002, 0.9, and 0.0, respectively. The iRadonMap is implemented on PyTorch deep learning framework. 10 The iRadonMap is trained for one week using two NVIDIA Tesla P40 graphics processing units (GPUs) with 24 GB memory capacity each.
About PubCompare
Our mission is to provide scientists with the largest repository of trustworthy protocols and intelligent analytical tools, thereby offering them extensive information to design robust protocols aimed at minimizing the risk of failures.
We believe that the most crucial aspect is to grant scientists access to a wide range of reliable sources and new useful tools that surpass human capabilities.
However, we trust in allowing scientists to determine how to construct their own protocols based on this information, as they are the experts in their field.
Ready to get started?
Sign up for free.
Registration takes 20 seconds.
Available from any computer
No download required
Revolutionizing how scientists
search and build protocols!