def eval_model(imgarray_evaldata, params):
for i in range(params["image_nr"]):
pc_utilities.process_messenger("EVALUATING IMAGE " + str(i))
pc_config.eval_img_id = i
# generate data set
pc_dataset.slice_image_hdf5(imgarray_image=imgarray_evaldata, imgarray_labels="none", sampletype="fullimage")
# segment the images
# using the model and the probability calibration function, we evaluate the classifier on a real image with unknown labels
y_truth, y_model = run_classifier(mode="probcalibratedregression", sampletype="fullimage")
y_model = y_model.reshape(
(len(y_model) ** (1.0 / 2), len(y_model) ** (1.0 / 2))
) # return a 2d map of the image
prob_map = y_model
print "prob map dimension", y_model.shape
if pc_config.save_truth:
y_truth = y_truth.reshape(
(len(y_truth) ** (1.0 / 2), len(y_truth) ** (1.0 / 2))
) # return a 2d map of the image
prob_map = np.concatenate([y_model, y_truth]) # save result and truth next to eachother
# save numpy
print "saving probability map to ", pc_config.fname_segmentation_pmap_npy
np.save(pc_config.fname_segmentation_pmap_npy, prob_map)
# postprocess the probability map to generate json and png data
postprocess_image(params)
# delete hdf5 samples that are no longer needed
pc_dataset.delete_data_hdf5()
pc_utilities.process_messenger("EVALUATION COMPLETE.")
def train_model(imgarray_traindata, imgarray_trainlabels, params):
# training
pc_dataset.slice_image_hdf5(imgarray_traindata, imgarray_trainlabels, sampletype="class01equal")
pc_utilities.process_messenger("MODEL TRAINING.")
training_loop()
# calibration
pc_utilities.process_messenger("MODEL TRAINING COMPLETE. NOW PROBABILITY CALIBRATION.")
pc_dataset.slice_image_hdf5(imgarray_traindata, imgarray_trainlabels, sampletype="random")
train_probability_calibration()
pc_dataset.delete_data_hdf5()
pc_utilities.process_messenger("PROBABILITY CALIBRATION COMPLETE. CLASSIFIER IS READY.")