def process_oneimage(self, image_file):
if not self.params.files['model_trash'] == 'None':
img, dummy = classifications.create_image(image_file,
cropped=False)
predicted_label, prob_trash = self.cnn_trash.classify(
img, char_sizes=None)
else:
prob_trash = 100
predicted_label = 1
if predicted_label == 0:
# TRASH goes to Recycle bin
predicted_type = self.params.type_dict_trash[str(predicted_label)]
prob_taxon = 0 # use np.NaN instead
self.text.insert(
tk.END, 'TRASH model result : ' + 'type : ' + predicted_type +
' ; prob : ' + str(prob_trash) + '\n')
self.text.see(tk.END)
else:
# NOT TRASH
img, char_sizes = classifications.create_image(image_file,
cropped=True)
predicted_label, prob_taxon = self.cnn_taxon.classify(
img, char_sizes=char_sizes)
predicted_type = self.params.type_dict_taxon[str(predicted_label)]
self.text.insert(
tk.END, 'TAXON model result : ' + 'type : ' + predicted_type +
' ; prob : ' + str(prob_taxon) + '\n')
self.text.see(tk.END)
return predicted_type, prob_trash, prob_taxon
def process(self, images2process_list_indir):
measure_dir = os.path.join(self.params.dirs['root'],
self.params.dirs['measurement'])
for fo in images2process_list_indir:
cur_dir = os.path.join(measure_dir, fo[0], fo[1])
self.text.insert(tk.END, 'visiting : ' + cur_dir + '\n')
res_folder1 = os.path.join(self.params.dirs['root'],
self.params.dirs['classification'],
self.params.dirs['results'], fo[0])
check_folder(folder=res_folder1, create=True)
res_folder = os.path.join(res_folder1, fo[1])
check_folder(folder=res_folder, create=True)
for image in fo[2]:
self.text.insert(tk.END, 'processing : ' + image + '\n')
image_file = os.path.join(measure_dir, fo[0], fo[1], image)
im = classifications.create_image(image_file, cropped=False)
predicted_label, prob = self.cnn.classify(im)
predicted_type = keysWithValue(self.params.type_dict,
str(predicted_label))[0]
self.text.insert(
tk.END, 'result : ' + 'type : ' + predicted_type +
' ; prob : ' + str(prob) + '\n')
self.text.see(tk.END)
class_folder = os.path.join(res_folder, predicted_type)
check_folder(folder=class_folder, create=True)
shutil.copy(image_file, os.path.join(class_folder, image))
file = open(os.path.join(cur_dir, 'MeasureSum.xml'), 'w')
file.close()
def process_oneimage(self, image_file):
# t=time.time()
# print(self.correct_RGBShift)
if not self.params.files['model_trash'] == 'None':
print(self.correct_RGBShift)
img, dummy = classifications.create_image(
image_file,
cropped=False,
correct_RGBShift=self.correct_RGBShift)
predicted_label_trash, prob_trash = self.cnn_trash.classify(
img, char_sizes=None
) # prob_trash is actually probability of object
else:
prob_trash = 100
predicted_label_trash = 1 # object!
if predicted_label_trash == 0: # classified as trash
# TRASH goes to Recycle bin
predicted_type = self.params.type_dict_trash[str(
predicted_label_trash)]
prob_taxon = 0 # use np.NaN instead
# self.text.insert(tk.END, 'TRASH model result : '+'type : '+predicted_type+' ; prob : '+str(prob_trash)+'\n')
# self.text.see(tk.END)
else:
# NOT TRASH
img, char_sizes = classifications.create_image(
image_file,
cropped=True,
correct_RGBShift=self.correct_RGBShift)
predicted_label, prob_taxon = self.cnn_taxon.classify(
img, char_sizes=char_sizes)
if prob_taxon < self.params.thresholds['trash_thresh']:
predicted_label_trash = 0 # TRASH
predicted_type = self.params.type_dict_trash[str(
predicted_label_trash)]
else:
predicted_type = self.params.type_dict_taxon[str(
predicted_label)]
# self.text.insert(tk.END, 'TAXON model result : '+'type : '+predicted_type+' ; prob : '+str(prob_taxon)+'\n')
# self.text.see(tk.END)
# print(t-time.time())
return predicted_type, prob_trash, prob_taxon
def process_oneimage(self, image_file, man_type):
predicted_label = None
predicted_type = 'Others.Others.Unsure'
final_type = None
predicted_strength = None
char_sizes = None
if self.save_cropped:
tmp_folder = os.path.join(os.path.curdir, 'tmp_crop')
check_folder(tmp_folder, create=True)
save_file = os.path.join(tmp_folder, os.path.basename(image_file))
category = 'cropped'
else:
save_file = ''
category = ''
img, char_sizes = classifications.create_image(
image_file,
cropped=True,
correct_RGBShift=self.correct_RGBShift,
save_file=save_file,
category=category)
if img is not None and char_sizes is not None:
if self.params.processing['use_neural'] == 'True':
predicted_label, predicted_strength = self.cnn_taxon.classify(
img, char_sizes=char_sizes)
predicted_type = self.params.type_dict_taxon[str(
predicted_label)]
else:
predicted_strength = 65535
predicted_type = man_type
final_type = predicted_type
# Do checks on threshold, shape and size
row = self.params.threshold_df_taxon.loc[
self.params.threshold_df_taxon['Taxon'] == predicted_type]
if not row.empty:
min_l = row['min_l'].values[0]
max_l = row['max_l'].values[0]
min_rate = row['min_rate'].values[0]
max_rate = row['max_rate'].values[0]
strength_tsh = row['strength_tsh'].values[0]
if char_sizes[0] > 0:
rate = char_sizes[1] / char_sizes[0]
else:
rate = -1
if predicted_strength < strength_tsh:
final_type = 'Others.Others.Unsure'
if char_sizes[0] < min_l: # char_sizes[0] larger axis length
final_type = 'Others.Others.InappropriateSize'
if char_sizes[0] > max_l:
final_type = 'Others.Others.InappropriateSize'
if rate < min_rate or rate > max_rate:
final_type = 'Others.Others.InappropriateShape'
# self.text.insert(tk.END, 'TAXON model result : '+'type : '+predicted_type+' ; strength : '+str(strength_str)+'\n')
# self.text.see(tk.END)
# print(t-time.time())
return predicted_type, final_type, predicted_strength, char_sizes
user='******'
output_base_dir=os.path.join(r'C:\Users',user,'OneDrive\WaterScope')
#output_base_dir=r'd:\DATA\WaterScope'
model_file=os.path.join(r'.\model','cnn_model_trash.dnn')
cnn=classifications.cnn_classification(model_file)
#df_db = pd.read_csv(db_file,delimiter=';')
type_dict={'Trash':'0','Object':'1'}
# LOAD MODEL
included_extenstions = ['*.jpg', '*.bmp', '*.png', '*.gif']
image_list_indir = []
for ext in included_extenstions:
image_list_indir.extend(glob.glob(os.path.join(db_image_dir, ext)))
for i, image_file in enumerate(image_list_indir):
#image_file=os.path.join(r'C:\Users\SzMike\OneDrive\WaterScope\db_images\0044094_DHM2.0.20130101.Measurement20160926_161935..20161024T115654-0010.png')
im = classifications.create_image(image_file,cropped=False)
predicted_label, prob = cnn.classify(im)
if predicted_label==1:
shutil.copy(image_file,os.path.join(db_binsel_dir,os.path.basename(image_file)))