conda env create -f ./p2p_source_code/environment.yml -n myenv python=3.6
conda activate myenv

or

conda create --name myenv python=3.6

conda activate myenv
pip install ./p2p_source_code/requirements.txt

The model checkpoints and prototypes can be downloaded here: https://drive.google.com/drive/folders/1iB0HXFzyboXhIBSzzwhbkNPmEJFImXCj

For our project, we have downloaded the datatets and parsed them in ./car_datasets:

• trainDir: car_datasets/train_augmented_less (augmented datatets from train datasets)
• projectDir: car_datasets/train (train datasets with augmented operation)
• testDir: car_datasets/test

• You need to update the trainDir, projectDir, testDir in args.yaml to match your datasets directory.
• For proto2proto student kd training, set the teacher path in p2p_source_code/Experiments/Resnet50_18_cars/kd_Resnet50_18/args.yaml: backbone.loadPath. Use the teacher model trained previously. For eg.

loadPath: p2p_source_code/Experiments/Resnet50_18_cars/teacher_Resnet50/org/models/teacher_checkpoint_model.pth

• In order to save the protopypes for epochs, you need set the dir_path in ./p2p_source_code/src/services/_recognition/protopnet_basic.py for root_dir_for_saving_prototypes.
  • teacher prototypes: root_dir_for_saving_prototypes='./prototypes/teacher_prototypes'
  • student prototypes: root_dir_for_saving_prototypes='./prototypes/student_prototypes' (both student and teacher on same place)
  • kd prototypes: root_dir_for_saving_prototypes='./prototypes/kd_student_prototypes' in in p2p_source_code/src/services/_kd/protopnet_kd.py

In our project, we have pre-setup the dir matching with our datasets directory, and loadPath for training kd model to use our teacher model , and prototypes dir_path.

• settingsConfig.backbone.loadPath need to be added or updated in order to match with the trained model you neeed to convert.
• for converting kd model, you have to access in kd_Resnet50_18/org/args.yaml and change the loadPath to link your trained kd checkpoint model.

settingsConfig.backbone.loadPath: p2p_source_code/Experiments/Resnet50_18_cars/teacher_Resnet50/org/models/kd_checkpoint_model.pth

Then you need to run for generaing the complete models: "teacher_Resnet50", "student_Resnet18", "kd_Resnet50_18" with following codes in notebook:

import convert
runName= "teacher_Resnet50"  
main.main(runName)

In our project, we have converted the models in converted_models.

Set model paths in Experiments/Resnet50_18_cars/eval_setting/args.yaml:

• Teacherbackbone.loadPath
• StudentBaselinebackbone.loadPath
• StudentKDbackbone.loadPath.

To run the evalution with following codes in notebook:

import main
runName="eval_setting"
main.main(runName)

The plotting of metrics of AAPs, AJS, we saved in evaluation_metrics_plotting.

Here is the citation of the proto2proto basecode paper:

@inproceedings{Keswani2022Proto2ProtoCY,
  title={Proto2Proto: Can you recognize the car, the way I do?},
  author={Monish Keswani and Sriranjani Ramakrishnan and Nishant Reddy and Vineeth N. Balasubramanian},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022)},
  eprint={2204.11830},
  archivePrefix={arXiv},
  year={2022}
}