24679-HW2-image-autogluon-predictor

Lanternfly Image Classifier

This model is trained to classify images of spotted lanternflies vs. non-lanternfly objects using AutoGluon Multimodal.
It leverages the dataset rlogh/lanternfly-images, which contains two splits:

• augmented: 336 labeled images with various transformations
• original: 33 raw images

The goal of this project is to demonstrate how tabular + image data can be processed with AutoGluon and pushed to Hugging Face Hub.

---

Model Details

• Framework: AutoGluon Multimodal
• Backend: PyTorch
• Task: Binary Image Classification (lanternfly vs. non-lanternfly)
• License: MIT

---

Dataset

• Name: rlogh/lanternfly-images
• Splits:
  • augmented: 336 images
  • original: 33 images
• Classes:
  • lanternfly (positive class)
  • other (negative class)

---

Training Setup

• Library: AutoGluon Multimodal
• Hardware: Google Colab GPU (Tesla T4)
• Augmentation: Used dataset-provided augmentations
• Hyperparameters: Default AutoGluon fit() config with limited training time

---

Evaluation

The model was evaluated on the dataset splits. Reported metrics include:

• Validation Accuracy: 76.47%

• Accuracy: 0.5455

• F1 Score: 0.3850

AI Usage

ChatGPT was used to draft and format this model card