Roboflow Object Detection Evaluation

To evaluate a model hosted on Roboflow, run the roboflow_example.py file with the following arguments:

python3 examples/roboflow_example.py --eval_data_path <path_to_eval_data> \
--roboflow_workspace_url <workspace_url> \
--roboflow_project_url <project_url> \
--roboflow_model_version <model_version>

Compare Roboflow Model Versions

You can also compare different versions of Roboflow models.

To do so, open the examples/compare_roboflow_models.py file and add the workspace URL, project URL, and versions for each model you want to compare. You should also set a confidence threshold for each model version that will be used to filter out low confidence predictions.

Then, run the compare_roboflow_models.py script:

python3 examples/compare_roboflow_models.py

This script will generate a PDF called roboflow_model_comparison.pdf that shows the precision, recall, f1 score, and confusion matrix for each model.