def compute_on_dir(self, dirpath, outdir):
"""
Computes the output of all images in a directory with 'im' in their name.
Args:
dirpath (string): path to the data folder
outdir (string): path to the output folder. Created if not existing.
Returns:
"""
if not os.path.isdir(outdir):
os.mkdir(outdir)
_, _, files = next(os.walk(dirpath))
dir_len = 0
for file in files:
if 'im' in file:
dir_len += 1
gen = BatchGenerator(traindir=dirpath,
city_model=self,
trainsetsize=dir_len,
batchsize=1,
traindirsize=dir_len)
counter = 0
inputs = gen.generate_input_only(option=self.prop_dict['trainset'][0])
for x in inputs:
if counter > dir_len:
break
x = np.squeeze(x,axis=0)
y = np.squeeze(self.compute_output(x),axis=0)
Im = Image.fromarray(x[:,:,0:3].astype('uint8'))
Out = Image.fromarray(y.astype('uint8'))
Im.save(os.path.join(outdir,'input_'+str(counter)+'_.png'))
Out.save((os.path.join(outdir,'output_'+str(counter)+'_.png')))
counter += 1
print(str(counter))
def evaluate_full(self, validation_dir, validation_option = '', logfile=None):
valdirsize = len([name for name in os.listdir(validation_dir) if os.path.isfile(os.path.join(validation_dir,name)) and 'im' in name])
val_gen = BatchGenerator(traindir=validation_dir,
city_model=self,
trainsetsize=valdirsize,
batchsize=1,
traindirsize=valdirsize)
counter = 0
gen = val_gen.generate_input_only(validation_option)
metrics = ['acc']
for met in Metrics.valid_metrics:
metrics.append(Metrics.create_metrics(met,self))