We should end up with a fully‑functional pipeline that takes a single test image, pre‑processes it, extracts global and local descriptors, compares against prototypical feature models and outputs the most likely flower species identified along with a similarity score and if a flower has been identified, determine whether it is in a healthy state.
To efficiently tackle the project, we divided the work into the following components:
- Pre-processing: Construct the training dataset and apply necessary image pre-processing techniques to prepare the data for modeling.
- Model Training: Develop and train one or more models using the processed data. This includes all code related to model architecture, training routines, and experimentation with different approaches.
- Performance Evaluation: Implement code to assess model performance, including metrics calculation, result analysis, and visualization of outcomes.
- Accuracy
- total accuracy
- accuracy per category
- Confusion matrix
- Mean processing time
Recommended (CMake):
cmake -S . -B build
cmake --build build -j4./build/flower_classifier Final_project_proposalClassification results from our test runs can be found under the results directory.
If you need to recompile the PDF report from the LaTeX source file, run make inside the docs folder:
cd docs/; make