def eval(self, data_in, **kwargs):
batch_size = kwargs.get('batch_size', 0)
print_str = kwargs.get('print_str', True)
need_imgs = kwargs.get('need_imgs', False)
print_all_results = kwargs.get(
'print_all_results', False) # to print each results and index
if type(data_in) is dict:
time_start = time.time()
eval_dict = self.model.eval(data_in, **kwargs)
time_cost = time.time() - time_start
if print_str:
print('Evaluation time cost is {:.1f}'.format(time_cost))
return eval_dict
data_provider = data_in
ndata = data_provider.size
m = ndata // batch_size
n = ndata % batch_size
eval_dict_str = {}
eval_dict_img = {}
time_start = time.time()
for _ in range(m):
data_dict = data_provider(batch_size)
sub_eval_dict_str, sub_eval_dict_img = self.model.eval(
data_dict, **kwargs)
eval_dict_str = U.dict_concat(eval_dict_str, sub_eval_dict_str)
if need_imgs:
eval_dict_img = U.dict_append(eval_dict_img, sub_eval_dict_img)
if n > 0:
sub_eval_dict_str, sub_eval_dict_img = self.model.eval(
data_provider(n), **kwargs)
eval_dict_str = U.dict_concat(eval_dict_str, sub_eval_dict_str)
if need_imgs:
eval_dict_img = U.dict_append(eval_dict_img, sub_eval_dict_img)
if print_str:
time_cost = time.time() - time_start
print('Evaluate {} data, time cost is {:.1f}'.format(
ndata, time_cost))
print(' {}'.format(U.dict_to_str(eval_dict_str)))
if print_all_results:
results_dict = {}
for key in eval_dict_str:
value = eval_dict_str.get(key)
if value.shape[0] == ndata:
results_dict[key] = value
for i in range(ndata):
print_str = "Picture Index: {}\t".format(i)
for key in results_dict:
print_str += "{}: ".format(key)
for k in results_dict[key][i]:
print_str += "{:.5f} ".format(k)
print_str += "\t"
print(print_str)
if need_imgs:
return eval_dict_str, eval_dict_img
else:
return eval_dict_str
def eval(self, data_in, **kwargs):
batch_size = kwargs.get('batch_size', 1)
print_str = kwargs.get('print_str', True)
need_imgs = kwargs.get('need_imgs', False)
max_save_batch = kwargs.get('max_save_batch', float('inf'))
if type(data_in) is dict:
time_start = time.time()
eval_dict = self.model.eval(data_in, **kwargs)
time_cost = time.time() - time_start
if print_str:
print('Evaluation time cost is {:.1f}'.format(time_cost))
return eval_dict
data_provider = data_in
ndata = data_provider.size
m = ndata // batch_size
n = ndata % batch_size
results = {}
imgs = {}
saved_batch_count = 0
if print_str:
print('Evaluating {} data...'.format(ndata), end='')
time_start = time.time()
for i in range(m):
sub_results = self.model.eval(data_provider(batch_size), **kwargs)
sub_imgs = sub_results.pop(
'imgs') if 'imgs' in sub_results else None
results = U.dict_concat(results, sub_results)
if need_imgs and sub_imgs is not None and saved_batch_count < max_save_batch:
imgs = U.dict_append(imgs, sub_imgs)
saved_batch_count += 1
if print_str:
print('\r\tEvalated {}/{} data'.format(batch_size * (i + 1),
ndata),
end='')
if n > 0:
sub_results = self.model.eval(data_provider(n), **kwargs)
sub_imgs = sub_results.pop(
'imgs') if 'imgs' in sub_results else None
results = U.dict_concat(results, sub_results)
if need_imgs and sub_imgs is not None and saved_batch_count < max_save_batch:
imgs = U.dict_append(imgs, sub_imgs)
saved_batch_count += 1
if print_str:
print('\r\tEvalated {}/{} data'.format(ndata, ndata))
time_cost = time.time() - time_start
print('Time cost is {:.1f}'.format(time_cost))
print(' {}'.format(U.dict_to_str(results)))
if need_imgs:
return results, imgs
else:
return results
def _load_temp_file(self, n):
assert self._temp_dir is not None, 'Temp dir is None'
assert os.path.exists(self._temp_dir), 'Can\'t find temp directory \'{}\''.format(self._temp_dir)
data_dict = {}
for _ in range(n):
temp_filename = self._file_list[self._cur_i]
sub_data_dict = np.load(temp_filename, allow_pickle='TRUE').item()
self._next_idx()
data_dict = U.dict_concat(data_dict, sub_data_dict)
return data_dict