This project presents a comparison of handwritten digit recognition performance between LabView and PyTorch, utilizing a Convolutional Neural Network (CNN) model.
Below is a detailed table comparing various metrics between LabVIEW and PyTorch implementations.
| Metric | LabView NI Value | PyTorch Value | Unit | Description |
|---|---|---|---|---|
| Model Accuracy | ||||
| Final Test Accuracy | 99.22 | 99.38 | % | Accuracy achieved on the test set after training. |
| Training Accuracy (at final epoch) | 99.39 | 99.21 | % | Accuracy on the training set at the final epoch. |
| Validation Accuracy (at final epoch) | 99.37 | 99.16 | % | Accuracy on the validation set at the final epoch. |
| Training Time | ||||
| Time per Epoch | 46.5 | 33.6 | Seconds/Epoch | Time taken to complete one epoch of training. |
| Total Training Time | 696 | 504 | Seconds | Total time taken to train the model over all epochs. |
| Average Time per Batch | 0.104 | 0.0716 | Seconds/Batch | Average time taken to process a single batch during training. |
| Memory Usage | ||||
| Maximum Memory Usage | 396.4 MB | 641.47 MB | MB or GB | Peak memory usage during training (RAM). |
| Average Memory Usage | 383 MB | 595.9 MB | MB or GB | Average memory consumption during training. |
| Inference Time | ||||
| Time per Batch (Inference) | 0.57 | 0.012685 | Seconds/Batch | Time taken to make predictions on a batch during inference. |
| Time per Sample (Inference) | 0.00253 | 0.000791 | Seconds/Sample | Average time taken to make a prediction on a single sample. |
| Resource Usage | ||||
| CPU Utilization During Training | 7 | 5.5 | % | Percentage of CPU usage during training. |
| CPU Utilization During Inference | 2 | 1.8 | % | Percentage of CPU usage during inference. |
| Model Convergence | ||||
| Number of Epochs to Convergence | 15 | 15 | Count | Number of epochs until the model converges to a stable accuracy level. |
| Final Training Loss | 0.00509 | 0.0079 | - | Final loss on the training set after training completes. |
| Final Validation Loss | 0.0249 | 0.0203 | - | Final loss on the validation set after training completes. |
The following plots demonstrate the change in model loss over each epoch during training.
