##################################################################################################

DATA ACCESS FORM:

Please fill out this form that asks for your name, a valid email address and the name of the institution you are affiliated with, to gain access to the data and model weights:

https://docs.google.com/forms/d/1FggY3gRfDUxH8D4hCH-OHwEIHqa2tSdFXwobvM27Qh4/edit

Thank you for your interest. The download link will be sent to your email once the form is completed. 

Please note: The data available through this link are of RGB images, each of shape [64 X 64 X 3]. If a higher resolution dataset is required, please address your request to: 

[email protected] (while cc'ing: [email protected])

Users are welcome to create "Issues" within this repository if they face any problems regarding execution and/or deployment of the trained model on their own data-sets or on the shared data: https://github.com/SCSLabISU/xPLNet/issues

##################################################################################################

LABELLING INFORMATION for shared data:

Please follow Figure 2 for class/label information for the soybean leaf image datset. Images within a particular folder fall under the class information associated with the folder name (number) as shown in Figure 2. For example, images in folder '0' correspond to images under the "Bacterial Blight" class and so on.

##################################################################################################

CITATION:

If you use this dataset and/or the methods proposed in our research please cite our PNAS paper available at:
http://www.pnas.org/content/115/18/4613

Bibtex:

@article {Ghosal4613,
	author = {Ghosal, Sambuddha and Blystone, David and Singh, Asheesh K. and Ganapathysubramanian, Baskar and Singh, Arti and Sarkar, Soumik},
	title = {An explainable deep machine vision framework for plant stress phenotyping},
	volume = {115},
	number = {18},
	pages = {4613--4618},
	year = {2018},
	doi = {10.1073/pnas.1716999115},
	publisher = {National Academy of Sciences},
	issn = {0027-8424},
	URL = {http://www.pnas.org/content/115/18/4613},
	eprint = {http://www.pnas.org/content/115/18/4613.full.pdf},
	journal = {Proceedings of the National Academy of Sciences}
}

##################################################################################################

THE FOLLOWING DESCRIBES THE MODEL ARCHITECTURE AND HYPER-PARAMETERS (use the seed # provided to reproduce the results from the paper):

seed = 1337 (use numpy.random.seed(seed))

Arch 1:
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 128, 62, 62)       3584      
_________________________________________________________________
batch_normalization_1 (Batch (None, 128, 62, 62)       248       
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 128, 60, 60)       147584    
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 128, 30, 30)       0         
_________________________________________________________________
batch_normalization_2 (Batch (None, 128, 30, 30)       120       
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 128, 28, 28)       147584    
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 128, 14, 14)       0         
_________________________________________________________________
batch_normalization_3 (Batch (None, 128, 14, 14)       56        
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 128, 12, 12)       147584    
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 128, 6, 6)         0         
_________________________________________________________________
batch_normalization_4 (Batch (None, 128, 6, 6)         24        
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 128, 4, 4)         147584    
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 128, 2, 2)         0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 512)               0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 500)               256500    
_________________________________________________________________
dropout_2 (Dropout)          (None, 500)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 100)               50100     
_________________________________________________________________
dropout_3 (Dropout)          (None, 100)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 9)                 909       
=================================================================
Total params: 901,877
Trainable params: 901,653
Non-trainable params: 224
_________________________________________________________________

OPTIMIZER USED: Adam (lr = 0.001; the default lr from "Adam: A Method for Stochastic Optimization" by Kingma et al.)
BATCH_SIZE for Training: 60
Epochs = 150 (90s per epoch)

INPUT SHAPE: (3, 64, 64) [channels first]
INPUT PRE-PROCESSING: Input to the network must be normalized by the maximum value of the input image channels before being fed to the trained network for online/offline classification. 
FRONTEND: Keras
BACKEND: Theano

GPU Used: NVIDIA GeForce GTX TITAN X (12207 MB of dedicated GPU memory) 
CUDA 8.0 (cuDNN 5.1)

Training Samples = 59184 (validation split = 0.1)
Test Samples = 6576

Test Accuracy = 94.13%