Releases: kspruthviraj/Plankiformer
For classifying zoo and phytoplankton
Bug fixes. Specifically to create_ensemble.py script where the sorting was done based on the class names which overestimated the performance.
Thresholding and updated the script
To predict, use the below command
python predict.py -test_path ./test_data/ -test_outpath ./out/phyto_out_2 / -main_param_path ./out/phyto_out_2/ -model_path ./out/phyto_out_2/trained_models/Init_0/ ./out/phyto_out_2/trained_models/Init_1/ ./out/phyto_out_2/trained_models/Init_2/ ./out/phyto_out_2/trained_models/Init_3/ ./out/phyto_out_2/trained_models/Init_4/ -ensemble 2 -finetuned 2 -threshold 0.0
Added thresholding to set while predicting
To predict, use the below command
python predict.py -test_path ./test_data/Phyto_test/ -model_path ./out/Zooformer_deploy/trained_models/Init_0/ ./out/Zooformer_deploy/trained_models/Init_1/ ./out/Zooformer_deploy/trained_models/Init_2/ -test_outpath ./out/Zooformer_deploy/test_ens_pred_3_models/ -main_param_path ./out/Zooformer_deploy/ -ensemble 2 -finetuned 2 -threshold 0.0
v1.1
Debugged the error that was saving the predictions in to the text file in a wrong way.
Transformer based model for plankton classification
Brief description on how to use the models for predictions on unseen data. For training the model, please look at the Readme.md
You can either use single model for making predictions or ensemble models (multiple models) to make predictions.
1) Single model prediction command:
python predict.py -test_path /test_data/test_images -model_path ./out/phyto/trained_models/Init_0/ -test_outpath ./out/on_test_result/ -main_param_path ./out/phyto/ -ensemble 0
2) Ensemble model prediction command:
python predict.py -test_path /test_data/test_images -model_path ./out/phyto/trained_models/Init_0/ ./out/phyto/trained_models/Init_1/ ./out/phyto/trained_models/Init2/ -test_outpath ./out/on_test_result/ -main_param_path ./out/phyto/ -ensemble 2
Explanation of the parameters:
test_path --> directory of images where you want to predict
model_path --> location of the trained model/s
test_outpath --> directory where you want to save the predicted results
main_param_path --> directory where the training parameters are saved
ensemble --> to choose either arithmetic (1) or geometric (2) or no ensemble (0)