Bring Your Own Model

This tool is used to leverage custom pre-trained PyTorch models for image classification tasks, providing support for fine-tuning and further optimization of these models within the PyTorch deep learning framework.

The following example demonstrates how Bring Your Own Model can be used for a fleet management use case. A car rental company wants to check that its customers return the correct car to the lot. With this application, the company can train a custom image classification model with PyTorch to identify a car's make, model, and year. Then, customers submit photos of their returned cars, and the custom model verifies the car information.

• Reads PyTorch model and synset files. The model file must be in TorchScript format, and the synset file must be a text file with one class label per line.
• Sends image files to model one at a time.
• Allows users to process an image before inference (i.e. resize, center crop, normalization of pixels).
• Provides flexibility to adjust the number of highest-ranked classes returned by model.
• Returns the classifications as a JSON string with class labels and their corresponding probabilities.
• Supports the addition of new machine learning applications (e.g. object detection, natural language processing, etc.)

If you have ideas for future improvements, please leave a comment.

Changed Appian version and updated dependency versions.