def predict(self, img_list=[], doubles_tol=0, edge_tol=0):
'''
output:
predictions: list of features
n images => n lists of features
'''
crop_px = self.estimator.pixels
yolo_predictions = self.localizer.predict(img_list=img_list)
yolo_predictions = nodoubles(yolo_predictions, tol=doubles_tol)
(imcols, imrows, channels) = img_list[0].shape
old_shape = (imrows, imcols)
yolo_predictions = no_edges(yolo_predictions,
tol=edge_tol,
image_shape=old_shape)
out_features, est_images, scales = crop_feature(
img_list=img_list, xy_preds=yolo_predictions, new_shape=crop_px)
structure = list(map(len, out_features))
char_predictions = self.estimator.predict(img_list=est_images,
scale_list=scales)
zpop = char_predictions['z_p']
apop = char_predictions['a_p']
npop = char_predictions['n_p']
for framenum in range(len(structure)):
listlen = structure[framenum]
frame = out_features[framenum]
index = 0
while listlen > index:
feature = frame[index]
feature.model.particle.z_p = zpop.pop(0)
feature.model.particle.a_p = apop.pop(0)
feature.model.particle.n_p = npop.pop(0)
feature.model.instrument = self.instrument
index += 1
return out_features
savedict = []
#just do one at a time for now
for i in range(numimgs):
filepath = path + '/image' + str(i).zfill(4) + '.png'
localim = cv2.imread(filepath)
localpreds = [x for x in MLpreds if x['framenum'] == i]
#reformat for cropping
for pred in localpreds:
localxy = {"conf": 1}
x_p = pred['x_p']
y_p = pred['y_p']
ext = pred['shape'][0]
localxy["bbox"] = [x_p, y_p, ext, ext]
features, _, _ = crop_feature(img_list=[localim], xy_preds=[[localxy]])
#instatiates a feature, puts in data, coords, x_p, y_p
if len(features[0]) != 1:
print('Something went wrong')
print(len(features[0]))
feature = features[0][0]
feature.deserialize(pred) #puts in rest of feature info
start = time()
result = feature.optimize(method='amoeba-lm')
print("Time to fit: {:03f}".format(time() - start))
print(result)
localdict = feature.serialize(exclude=['data'])
localdict['framenum'] = i
localdict['framepath'] = os.path.abspath(filepath)
localdict['redchi'] = result.redchi
savedict.append(localdict)
def report_feature(self, conditions, predtype, crop_dir=None):
self.do_omit()
if predtype == 'ML':
predictions = [
x for x in self.ML_preds for cond in conditions
if np.all(cond(x))
]
elif predtype == 'refined':
predictions = [
x for x in self.refined_preds for cond in conditions
if np.all(cond(x))
]
else:
print('Invalid predictions type')
for pred in predictions:
localim = cv2.imread(pred['framepath'])
shape = pred['shape']
localxy = {"conf": 1}
x_p = pred['x_p']
y_p = pred['y_p']
ext = shape[0]
localxy["bbox"] = [x_p, y_p, ext, ext]
features, _, _ = crop_feature(img_list=[localim],
xy_preds=[[localxy]])
feature = features[0][0]
feature.deserialize(pred)
h = feature.model.hologram()
res = feature.residuals()
print(pred)
fig, axes = plt.subplots(1, 3)
for ax in axes:
ax.set_xticks([])
ax.set_yticks([])
ax.xaxis.set_label_position('top')
(ax1, ax2, ax3) = axes
cropped_data = np.clip(feature.data.reshape(shape) * 100, 0, 255)
if crop_dir is not None:
bad_name = True
counter = 0
while bad_name:
crop_path = os.path.abspath(
crop_dir) + '/frame{}_img{}.png'.format(
pred['framenum'], counter)
if not os.path.isfile(crop_path):
bad_name = False
else:
counter += 1
cv2.imwrite(crop_path, cropped_data)
print(crop_path)
ax1.imshow(cropped_data, cmap='gray')
ax2.imshow(np.clip(h.reshape(shape) * 100, 0, 255), cmap='gray')
ax3.imshow(res.reshape(shape), cmap='gray')
ax1.set_xlabel('Data')
ax2.set_xlabel('Predicted Hologram')
ax3.set_xlabel('Residual')
z_str = '%.3f' % pred['z_p']
a_str = '%.3f' % pred['a_p']
n_str = '%.3f' % pred['n_p']
fig.suptitle('Frame {}'.format(pred['framenum']))
fig.text(0.5,
0.2,
'z_p = {}px \n a_p = {}um \n n_p = {}'.format(
z_str, a_str, n_str),
horizontalalignment='center')
plt.show()