modelName = 'FerretAndMouseROIs_Relu_BatchNorm_Dropout{}_{}'.format(
options['blocks'], options['epochs']) #_Relu_5x100_BatchNorm
filePath_model = r'{}\Code\Dave\{}_model_{}.h5'.format(
baseFolder, modelType, modelName)
filePath_history = r'{}\Code\Dave\{}_history_{}.hdf5'.format(
baseFolder, modelType, modelName)
filePath_params = r'{}\Code\Dave\{}_params_{}.txt'.format(
baseFolder, modelType, modelName)
if reuseModel:
model.loadModel(filePath_model)
model.loadModelHistory(filePath_history)
model.loadModelParameters(filePath_params)
if trainModel:
model.trainModel(train_images, train_labels, test_images, test_labels,
options)
model.saveModel(filePath_model)
model.saveModelHistory(filePath_history)
model.saveModelParameters(filePath_params)
# Show training/test accuracy and loss for model
try:
modelHistory = model.modelHistory
metrics = modelHistory.keys()
for metric_name in metrics:
metric = modelHistory[metric_name]
fig, ax = plt.subplots()
ax.plot(metric.T)
if (metric_name.startswith('accuracy')):
ax.set_ylim([0.5, 1])
elif (metric_name.startswith('loss')):
ax.set_ylim([0, 1.25 * np.percentile(metric.flatten(), 95)])
options.update({'augment_flipHoriz':
True}) # Boolean flag for horizontal flip
options.update({'augment_flipVert':
True}) # Boolean flag for vertical flip
options.update(
{'augment_fillMode': 'reflect'}
) # Points outside image boundary are filled with: {"constant", "nearest", "reflect" or "wrap"}
options.update(
{'augment_fillVal': 0}
) # If augment_fillMode=='constant', points outside image boundary are filled with fillVal
options.update({'augment_zca_whitening': False}) # Apply ZCA whitening
model = UNet()
if trainModel:
accuracy = model.trainModel(train_images, train_labels, test_images,
test_labels, options)
model.saveModel(r'D:\Code\ROI Segmentation\Code\Dave\Unet_model2.h5')
# Show accuracy model
fig, ax = plt.subplots()
ax.plot(accuracy)
ax.set_ylim([0.5, 1])
ax.set_ylabel('Classification accuracy')
ax.set_xlabel('Training epoch')
ax.legend(['Training data', 'Test data'])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
else:
model.loadModel(r'D:\Code\ROI Segmentation\Code\Dave\Unet_model2.h5')
# Get validation experience (ferret image and ROIs)