Gain insights into image classification with PyTorch. Learn about data preprocessing, model training, fine-tuning, and deploying models using ONNX for real-world applications.

Introduction to PyTorch image classification-01.png

wfngenv.tar.gz

wfngenv

terminalenv

PyTorch is a machine learning framework used in a wide array of popular applications, including Tesla’s Autopilot and Pyro, Uber’s probabilistic modeling engine.

This course is an introduction to image classification using PyTorch’s computer vision models for training and tuning your own model. You’ll start with the fundamental concepts of applying machine learning and its applications to image classification before exploring the process of training your AI model. You’ll prepare data for intake by the computer vision model with image pre-processing, set up pipelines for training your model, and fine-tune the variables to improve predictive performance. You’ll finish by deploying the image classification model by converting to ONNX format and serving it via REST API.

By the end of this course, you’ll be able to build and deploy your own image classification models from scratch.

Getting Started with Image Classification with PyTorch

**Asymmetric loss** is different from standard loss functions. It operates differently on both positive and negative samples.

It’s extremely useful as a loss function for multi-label classification tasks. We can classify an image into a few positive labels and many negative labels, which will result in a positive-negative imbalance situation, making it harder to optimize.

Asymmetric loss mitigates this issue by dynamically down-weighting negative samples with high confidence. It’ll discard any potential mislabeled samples. It’s easy to implement without any increase in training time or complexity.
The PyTorch Image Model provides two different asymmetric losses:

* `AsymmetricLossSingleLabel`
* `AsymmetricLossMultiLabel`

## The `AsymmetricLossSingleLabel` module
The `AsymmetricLossSingleLabel` works best for problems that only require a single label as the prediction.

Experiment with the interactive playgrounds below to learn more.


**Asymmetric loss** is different from standard loss functions. It operates differently on both positive and negative samples.

It’s extremely useful as a loss function for multi-label classification tasks. We can classify an image into a few positive labels and many negative labels, which will result in a positive-negative imbalance situation, making it harder to optimize.

Asymmetric loss mitigates this issue by dynamically down-weighting negative samples with high confidence. It’ll discard any potential mislabeled samples. It’s easy to implement without any increase in training time or complexity.
The PyTorch Image Model provides two different asymmetric losses:

* `AsymmetricLossSingleLabel`
* `AsymmetricLossMultiLabel`

# The `AsymmetricLossSingleLabel` module
The `AsymmetricLossSingleLabel` works best for problems that only require a single label as the prediction.

Experiment with the interactive playgrounds below to learn more.


Learn about asymmetric loss for both single-label and multi-label classification tasks.

Introduction

Basic Concepts

Augmentation

Loss

Training

Model Conversion

Deployment

Appendix

Conclusion

Asymmetric Loss

The `AsymmetricLossSingleLabel` module