Problem Statement
1. Problem Overview
The objective is to develop an automated Image Classification system using Convolutional Neural Networks (CNNs) or Transfer Learning. The system must effectively extract spatial features to classify insect imagery across 75 distinct species, simulating real-world biodiversity monitoring tasks.
2. Training Data
The dataset contains over 1,000+ images capturing various butterfly species:
-Images (train & test folders): Raw RGB images. Each image belongs to a single species.
-Training Labels (Training_set.csv): Maps training images (filename) to their ground-truth biological classes (label).
-Testing Set (Testing_set.csv): Contains filenames in the test folder. This is the unlabelled evaluation set you need to predict.
3. Problem Requirements
Input: A raw RGB image file of a single butterfly.
Output: A text label representing the predicted butterfly category (out of 75 classes).
Available Packages: Default Google Colab packages and standard Vision libraries (e.g., torchvision, keras.applications).
Internet Access: Completely Offline. External datasets or API-based inferences are strictly prohibited.
4. Evaluation Metrics
Accuracy: The percentage of correctly predicted observations over the total unseen test set. This single metric evaluates the model's precision in distinguishing fine-grained biological classes.
License: CC0. Public Domain.