Look requirements.txt
Plant village dataset with few extra classes of images
link: https://drive.google.com/open?id=1EuePtPjB2N_tZlaHjUTcw4p1JR4RldjP
- store the images in folder datasets/raw/species/disease-type
- run segment.py inside preprocessing/image-segmentation
python segment.py
python chk_move_segmented_image.py
**note: the segmentation_verification model is to be trained first which uses transfer learning in mobilenetV2 model.
python segmentation_verification_model_train.py
python resume_training.py
(to train for more epochs or updated data)
python dataSplit.py --source original/augmented
-
move to folder trainer
-
edit CNN models inside models
-
edit train_diseases.py and train_species.py for using the required models
python train_diseases.py -m 1
trains first group of models 5.
python train_diseases.py -m 2
trains second group of models
- Similarly for species.
note: The training is always resumed so use init_disease.py .... inside progress to restart training session
Use tensorboard to visualize events inside the logs/{} Or use pickle files
tensorboard --logdir = "logs/{}"
Choose best models with best modelScore = (max.acc+max.val_acc+max.f1_score+max.val_f1_score)/4 + log20(2000000/params)
convert to protobuf format remove unnecessary ops and save for inference
- move to finalise
python finalise.py
python deploy.py
Automatic Leaf Extraction from Outdoor Images
https://www.researchgate.net/publication/320649606_Plant_disease_identification_A_comparative_study
Krishna Upadhyay (krishnaupadhyay1997@gmail.com)
Sanjay Karki (sonJ9)
Simon Dahal (simonsd054@gmail.com)
Yogesh Rai (ygsh.spcry5@gmail.com)