def _test_infer(self):
infer_args = {
'--cuda': True,
'--model-class': self.args['--model-class'],
'--seed': "0",
'--beam-size': "2",
'--max-dec-step': "50",
'--beam-class': "models.beam.Beam",
'--model-type': "generator",
'--batch-size': 3,
'MODEL_PATH': os.path.join(self.args['--save-to']),
'TEST_SET_FILE': os.path.join(self.dataset_prefix, "test.jsonl"),
'OUTPUT_FILE': os.path.join(self.dataset_prefix, "result.json")
}
infer = Infer(infer_args)
hypos = infer.infer()
return hypos
# 数据路径
data_path = "trainPart/dataset/"
trainList = './trainPart/train_data.txt'
testList = './trainPart/test_data.txt'
readPath = "./trainPart/dataset/IMG/"
savePath = "./trainPart/dataset/IMG_Seg/"
if not os.path.exists(savePath):
os.mkdir(savePath)
# 多少比例用作训练集
ratio = 0.8
window = ImageGrab.grab() # 获得当前屏幕,存窗口大小
imm = cv2.cvtColor(np.array(window), cv2.COLOR_RGB2BGR) # 转为opencv的BGR格式
width, height = window.size
r = Infer(width, height, imm)
inf = Infer(width, height, imm)
inf.infer_pictures(readPath, savePath)
# 只读csv文件
with open(data_path + "log.txt", 'r') as logFile:
_list = logFile.readlines()
# 判断图片数是否匹配
ls_imgs = glob.glob(data_path + 'IMG/*.jpg')
print(len(ls_imgs))
print(len(_list))
assert len(ls_imgs) == len(_list), 'number of images does not match'
if (os.path.exists(trainList)):
os.remove(trainList)
from infer import Infer
from config import ModelBasic, TrainBasic
infer = Infer(corpus_name=TrainBasic.dataset,
run_name=TrainBasic.runname,
sample_num=TrainBasic.batch_size,
sample_dim=ModelBasic.in_out_dim)
infer.generate(0, 0)
from infer import Infer
import time
rect = (678, 350, 1078, 550)
j = pyvjoy.VJoyDevice(1)
place = fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
[infer_program, feeded_var_names,
target_var] = fluid.io.load_inference_model(dirname="./model_infer/",
executor=exe)
window = ImageGrab.grab() # 获得当前屏幕,存窗口大小
img = cv2.cvtColor(np.array(window), cv2.COLOR_RGB2BGR) # 转为opencv的BGR格式
width, height = window.size
r = Infer(width, height, img)
inf = Infer(width, height, img)
segFlag = False # 是否是处理分割的图片
def control(ang, brake):
if ang > 60:
ang = 60
if ang < -60:
ang = -60
global j
x = ang / 180 + 0.5
j.data.wAxisX = int(x * 32767)
j.data.wAxisY = int(brake * 32767)
j.data.wAxisZ = 0
def evaluate(self, epoch, step):
infer = Infer(corpus_name=self.corpus_name, run_name=self.run_name, sample_num=TrainBasic.eval_size, sample_dim=ModelBasic.in_out_dim)
infer.generate(epoch, step)
args.feature_shape = tuple([int(x) for x in args.feature_shape.split(',')])
args.record_shape = tuple([int(x) for x in args.record_shape.split(',')])
task = args.__dict__['crohns_or_polyps']
if args.pytorch:
trainer = PytorchTrainer(args)
trainer.train()
else:
# if task == 'Polyps_CT':
# decode_record = generate_decode_function(args.record_shape, 'image')
# model = VGG
if task == 'Crohns_MRI':
decode_record = generate_decode_function(args.record_shape,
'axial_t2')
model = ResNet3D
args.__dict__['decode_record'] = decode_record
if args.mode == 'train':
trainer = Trainer(args, model)
trainer.train()
elif args.mode == 'test':
infer = Infer(args, model)
infer.test(os.path.join(args.base, args.test_datapath))
axial_path = '/vol/bitbucket/rh2515/MRI_Crohns/A/A36 Axial T2.nii'
coords = [198, 134, 31]
infer.infer(axial_path, coords, args.record_shape,
args.feature_shape)
# model_path_list.append(os.path.join(directory, filename))
model_path_list.append(filename)
print(model_path_list)
model_path_list.sort(key=lambda x:int(x.split('_')[2]))
# model_path_list.sort()
for i in range(len(model_path_list)):
if i < 29 or i > 49:
continue
if i % opts.span == opts.span-1:
# if i > -1:
print(model_path_list[i])
model_state = torch.load(os.path.join(directory, model_path_list[i]))
Model.load_state_dict(model_state)
Model.to("cuda")
summary(Model, (3, 64, 64))
break
dataset = Dataload(imgpath=opts.image_folder, csv_name=csv_name)
with torch.no_grad():
print("infer")
kld_list = Infer(Model, dataset, batch_size=opts.bs, latent_dim=opts.latent_dim, output_folder=opts.model_folder)
index_of_change = linearSearch(kld_list, opts.top_K)
np.save(os.path.join(opts.model_folder, 'scene_change_res' + str(i) + '.npy'), index_of_change)
import time
segFlag = False # 是否是处理分割的图片
rect = (678, 350, 1078, 550)
j = pyvjoy.VJoyDevice(1)
place = fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
[infer_program, feeded_var_names,
target_var] = fluid.io.load_inference_model(dirname="./model_infer/",
executor=exe)
window = ImageGrab.grab() # 获得当前屏幕,存窗口大小
img = cv2.cvtColor(np.array(window), cv2.COLOR_RGB2BGR) # 转为opencv的BGR格式
width, height = window.size
inf = Infer(width, height, img) if segFlag else None
def control(ang, brake):
if ang > 60:
ang = 60
if ang < -60:
ang = -60
global j
x = ang / 180 + 0.5
j.data.wAxisX = int(x * 32767)
j.data.wAxisY = int(0.2 * 32767)
j.data.wAxisZ = 0
j.update()
print("Running with arguments: ")
for a in args.__dict__:
print(str(a) + ": " + str(args.__dict__[a]))
args.attention = int(args.attention)
# args.localisation = int(args.localisation)
# args.mixedAttention = int(args.mixedAttention)
# args.deeper = int(args.deeper)
args.feature_shape = tuple([int(x) for x in args.feature_shape.split(',')])
args.record_shape = tuple([int(x) for x in args.record_shape.split(',')])
task = args.__dict__['crohns_or_polyps']
# if task == 'Polyps_CT':
# decode_record = generate_decode_function(args.record_shape, 'image')
# model = VGG
if task == 'Crohns_MRI':
decode_record = generate_decode_function(args.record_shape, 'axial_t2')
model = ResNet3D
args.__dict__['decode_record'] = decode_record
if args.mode == 'train':
trainer = Trainer(args, model)
trainer.train()
elif args.mode == 'test':
infer = Infer(args, model)
# the following are harded coded examples (change to run inference on other images)
axial_path = './examples/A1 Axial T2.nii'
coords = [198, 134, 31]
infer.infer(axial_path, coords, args.record_shape, args.feature_shape)
import itertools
config_file = json.load(open('./config.json'))
num_inferences = 1 # number of iteration for each configuration
for _ in range(num_inferences):
pass
for c in itertools.product(
config_file['n_layers'], config_file['n_filters'],
config_file['batch_size'], config_file['input_size'],
config_file['n_classes'], config_file['kernel_size'],
config_file['fc_units'], config_file['kernel_stride']):
config = {
'n_layers': c[0],
'n_filters': c[1],
'batch_size': c[2],
'input_size': c[3],
'n_classes': c[4],
'kernel_size': c[5],
'fc_units': c[6],
'kernel_stride': c[7]
}
print(config)
sl = Infer(config)
sl.get_data()
sl.model()
sl.loss()
sl.optimizer()
sl.infer()
del sl
#!/usr/bin/env python
''' Testing Infer on HF Support KB. '''
from infer import Infer
if __name__ == "__main__":
kb_topics = [topic.rstrip('\n') for topic in open('./kb.txt')]
stoplist = set('for a of the and to in is are to how do can I ?'.split())
infer = Infer()
infer.build(kb_topics, stoplist, update=False, num_topics=len(kb_topics))
sims = infer.infer("How do I subscribe to a ticket")
print kb_topics[sims[0][0]]
num_inferences = 1 # number of iteration for each configuration
store = True
count = 0
for c in itertools.product(config_file['n_layers'], config_file['n_filters'], config_file['batch_size'],
config_file['input_size'],config_file['n_classes'], config_file['kernel_size'],config_file['fc_units'],
config_file['kernel_stride']):
print('configuration', count)
count += 1
for i in range(num_inferences):
print(i)
config = {'n_layers' : c[0],
'n_filters' : c[1],
'batch_size' : c[2],
'input_size' : c[3],
'n_classes' : c[4],
'kernel_size' : c[5],
'fc_units' : c[6],
'kernel_stride' : c[7]}
print(config)
sl = Infer(config)
sl.get_data()
sl.model()
sl.loss()
sl.optimizer()
model_complexity = sl.infer()
model_complexity['config'] = config
if store:
collection.insert(model_complexity)
del sl
result_file_name = image_file + '.txt'
with codecs.open(result_file_name, 'w', encoding='utf-8') as f:
for i, image_name in enumerate(image_name_list):
image_path = os.path.join(image_file, image_name)
try:
image = Image.open(image_path)
predict_text = ocr_engine.predict(image, long_info=False)
except:
predict_text = ''
print(image_path)
print(predict_text)
f.write('{}\t{}\n'.format(image_path, predict_text))
f.flush()
ocr_engine = Infer('/home/huluwa/tf_crnn/model/ctc_center')
if __name__ == "__main__":
TEST_OCR_MODEL = False
TEST_BATCH_OCR_MODEL = True
if TEST_OCR_MODEL:
root_dir = './data_example/test_data/xingjin'
gt_file = './data_example/test_data/xingjin1'
report_file = './testset_result_local.txt'
start_time = time.time()
test_ocr_model(root_dir, gt_file, report_file)
print('total cost time is %.4f ms' % ((time.time() - start_time) * 1000))
exit()
if TEST_BATCH_OCR_MODEL:
root_dir = './data_example/test_data/xingjin'
gt_file = './data_example/test_data/xingjin1'
report_file = './testset_result_batch.txt'
FLAGS.num_classes, FLAGS.embedding_dim,
len(data_processor.char2idx), FLAGS.hidden_size,
FLAGS.learning_rate)
elif args.model == 'bimpm':
model = BIMPM()
elif args.model == 'abcnn':
model = ABCnn()
elif args.task == "chatbot":
if args.model == "seq2seq_att":
word2inx = data_processor.char2idx
word2inx['<GO>'] = len(word2inx) + 1
word2inx['<EOS>'] = len(word2inx) + 1
model = Seq2SeqWithAtt(
FLAGS.max_len,
len(word2inx), # FLAGS.vocab_size,
word2inx, # FLAGS.word2inx,
FLAGS.embedding_dim,
FLAGS.state_size,
FLAGS.num_layers,
FLAGS.use_attention,
FLAGS.use_teacher_forcing,
FLAGS.learning_rate,
FLAGS.beam_width)
if args.mode == "train":
trainer = Train(model, FLAGS, sess_config, field_len=field_len)
trainer.train(sess, train_data, eval_data, test_data)
elif args.mode == "eval":
inferor = Infer(model, FLAGS, sess)
inferor.infer(eval_data, field_len=field_len)