def _initConfig(self):
self.name = 'mnist_original'
util.create_dir(DATASET_DIR)
self.h = INPUT_SHAPE[0]
self.w = INPUT_SHAPE[1]
self.c = INPUT_SHAPE[2]
self.num_classes = 10
def _initConfig(self):
self.name = 'mnist_original'
util.create_dir(DATASET_DIR)
self.h = INPUT_SHAPE[0]
self.w = INPUT_SHAPE[1]
self.c = INPUT_SHAPE[2]
self.num_classes = 10
def _initConfig(self):
self.name = 'cifar'
util.create_dir(DATASET_DIR)
self.frames = INPUT_SHAPE[0]
self.h = INPUT_SHAPE[1]
self.w = INPUT_SHAPE[2]
self.c = INPUT_SHAPE[3]
self.num_classes = 100
def _initConfig(self):
self.name = 'lrw'
util.create_dir(DATASET_DIR)
self.frames = INPUT_SHAPE[0]
self.h = INPUT_SHAPE[1]
self.w = INPUT_SHAPE[2]
self.c = INPUT_SHAPE[3]
self.num_classes = 500
def start_mnist_original_writer(self):
util.create_dir(MNIST_ORIGINAL_DATASET_DIR)
util.create_dir(MNIST_ORIGINAL_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(MNIST_ORIGINAL_TFRECORDS_DIR, 'mnist_classmap.txt')
if not util.isDirOrFileExist(classmapfile):
mnistClasses = util.mnistClassmap(numAsKeys=False)
util.writeClassmapFile(classmapfile, mnistClasses)
mnistReader = MnistReader()
for datasetType in dataset_types:
self.mnist_write_records(mnistReader, MNIST_ORIGINAL_TFRECORDS_DIR, 'mnist_original', datasetType)
def mnist_write_records(self, mnistReader, tfrecordsdir, datasetname, datasetType):
print("Creating dir for saving tf records!")
dirpath = util.create_dir(os.path.join(tfrecordsdir, datasetType))
print("Starting processing {} images".format(datasetType))
batchSize = WRITE_BATCH_SIZE
totalBatches = int(mnistReader.getNumOfExamples(datasetType) / batchSize) + 1
for i in range(totalBatches):
print("Writing batch {}/{}".format(i + 1, totalBatches))
currentBatch = mnistReader.getData(datasetType)
images = currentBatch[0][i*batchSize:(i+1)*batchSize].astype(dtype=np.float32)
labels = currentBatch[1][i*batchSize:(i+1)*batchSize]
if (datasetname == 'mnist'):
images = self.resize_images_mnist(images)
labels = np.argmax(labels, axis=1)
batch_filename = os.path.join(dirpath, '{}_{}_batch_{}.{}'.format(datasetname, datasetType, i, "tfrecords"))
self.writeMNIST(batch_filename, images, labels)
def cifar_write_records(self, cifarReader, tfrecordsdir, datasetname,
datasetType):
print("Creating dir for saving tf records!")
dirpath = util.create_dir(os.path.join(tfrecordsdir, datasetType))
print("Starting processing {} images".format(datasetType))
batchSize = WRITE_BATCH_SIZE
totalBatches = int(
cifarReader.getNumOfExamples(datasetType) / batchSize) + 1
for i in range(totalBatches):
print("Writing batch {}/{}".format(i + 1, totalBatches))
currentBatch = cifarReader.getData(datasetType)
images = currentBatch[0][i * batchSize:(i + 1) *
batchSize].astype(dtype=np.float32)
images = self.resize_images_cifar(images)
labels = currentBatch[1][i * batchSize:(i + 1) * batchSize]
batch_filename = os.path.join(
dirpath, '{}_{}_batch_{}.{}'.format(datasetname, datasetType,
i, "tfrecords"))
self.writeCIFAR(batch_filename, images, labels)
def start_cifar_writer(self):
util.create_dir(CIFAR_DATASET_DIR)
util.create_dir(CIFAR_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(CIFAR_TFRECORDS_DIR, 'cifar_classmap.txt')
cifarClassesFile = os.path.join(CIFAR_DATASET_DIR, 'cifar_classes.txt')
if not util.isDirOrFileExist(classmapfile):
cifarClasses = util.cifarClassmap(cifarClassesFile)
util.writeClassmapFile(classmapfile, cifarClasses)
cifarReader = CifarReader()
for datasetType in dataset_types:
self.cifar_write_records(cifarReader, CIFAR_TFRECORDS_DIR, 'cifar', datasetType)
def start_mnist_writer(self):
util.create_dir(MNIST_DATASET_DIR)
util.create_dir(MNIST_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(MNIST_TFRECORDS_DIR, 'mnist_classmap.txt')
if not util.isDirOrFileExist(classmapfile):
mnistClasses = util.mnistClassmap(numAsKeys=False)
util.writeClassmapFile(classmapfile, mnistClasses)
mnistReader = MnistReader()
for datasetType in dataset_types:
self.mnist_write_records(mnistReader, MNIST_TFRECORDS_DIR, 'mnist',
datasetType)
def start_cifar_writer(self):
util.create_dir(CIFAR_DATASET_DIR)
util.create_dir(CIFAR_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(CIFAR_TFRECORDS_DIR, 'cifar_classmap.txt')
cifarClassesFile = os.path.join(CIFAR_DATASET_DIR, 'cifar_classes.txt')
if not util.isDirOrFileExist(classmapfile):
cifarClasses = util.cifarClassmap(cifarClassesFile)
util.writeClassmapFile(classmapfile, cifarClasses)
cifarReader = CifarReader()
for datasetType in dataset_types:
self.cifar_write_records(cifarReader, CIFAR_TFRECORDS_DIR, 'cifar',
datasetType)
def start_lrw_writer(self):
util.create_dir(LRW_DATASET_DIR)
util.create_dir(LRW_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(LRW_TFRECORDS_DIR, 'lrw_classmap.txt')
if util.isDirOrFileExist(classmapfile):
lrwClasses = util.readClassmapFile(classmapfile, numsAsKeys=False)
else:
lrwClasses = util.lrwWordsToNumbers(LRW_DATASET_DIR)
util.writeClassmapFile(classmapfile, lrwClasses)
lrwReader = LRWReader()
for datasetType in dataset_types:
lrwMap = lrwReader.readLRWtoMap(LRW_DATASET_DIR, datasetType)
self.lrw_write_records(lrwMap, lrwClasses, 'lrw', datasetType)
def start_lrw_writer(self):
util.create_dir(LRW_DATASET_DIR)
util.create_dir(LRW_TFRECORDS_DIR)
dataset_types = config.config['dataset_types']
classmapfile = os.path.join(LRW_TFRECORDS_DIR, 'lrw_classmap.txt')
if util.isDirOrFileExist(classmapfile):
lrwClasses = util.readClassmapFile(classmapfile, numsAsKeys=False)
else:
lrwClasses = util.lrwWordsToNumbers(LRW_DATASET_DIR)
util.writeClassmapFile(classmapfile, lrwClasses)
lrwReader = LRWReader()
for datasetType in dataset_types:
lrwMap = lrwReader.readLRWtoMap(LRW_DATASET_DIR, datasetType)
self.lrw_write_records(lrwMap, lrwClasses, 'lrw', datasetType)
def lrw_write_records(self, videoMap, videoClassMap, datasetname,
dataset_type):
print("Creating dir for saving tf records!")
dirpath = util.create_dir(os.path.join(LRW_TFRECORDS_DIR,
dataset_type))
print("Clearing already written items!")
videoMap = self.clear_already_written(dirpath, videoMap)
print("Shuffling data!")
videoList = list(videoMap.items())
shuffle(videoList)
print("Starting processing {} images".format(dataset_type))
batchSize = WRITE_BATCH_SIZE
batchIndexer = self.get_number_of_written_tfrecords(dirpath)
totalBatches = int(len(videoList) / batchSize) + 1
batchCounter = 0
while len(videoList) > 0:
print("Writing batch {}/{}".format(batchCounter + 1, totalBatches))
currentBatch = videoList[:batchSize]
if len(videoList) <= batchSize:
videoList = []
else:
videoList = videoList[batchSize:]
batchWithImagesList = []
for video in currentBatch:
resizedVideos = len(batchWithImagesList)
if not resizedVideos % 100:
print("Videos read and resized: {}/{}".format(
resizedVideos, len(currentBatch)))
images = util.video_to_images(video[0])
images = self.resize_images_lrw(images)
batchWithImagesList.append(
(video[0], video[1], videoClassMap[video[1]], images))
batch_filename = os.path.join(
dirpath, '{}_{}_batch_{}.{}'.format(datasetname, dataset_type,
batchIndexer, "tfrecords"))
self.writeLRW(batch_filename, batchWithImagesList, dataset_type)
batch_stat_filename = os.path.join(
dirpath, '{}_{}_batch_{}.{}'.format(datasetname, dataset_type,
batchIndexer, "txt"))
self.write_written_images(batch_stat_filename, batchWithImagesList)
batchIndexer += 1
batchCounter += 1
def lrw_write_records(self, videoMap, videoClassMap, datasetname, dataset_type):
print("Creating dir for saving tf records!")
dirpath = util.create_dir(os.path.join(LRW_TFRECORDS_DIR, dataset_type))
print("Clearing already written items!")
videoMap = self.clear_already_written(dirpath, videoMap)
print("Shuffling data!")
videoList = list(videoMap.items())
shuffle(videoList)
print("Starting processing {} images".format(dataset_type))
batchSize = WRITE_BATCH_SIZE
batchIndexer = self.get_number_of_written_tfrecords(dirpath)
totalBatches = int(len(videoList)/batchSize) + 1
batchCounter = 0
while len(videoList) > 0:
print("Writing batch {}/{}".format(batchCounter + 1, totalBatches))
currentBatch = videoList[:batchSize]
if len(videoList) <= batchSize:
videoList = []
else:
videoList = videoList[batchSize:]
batchWithImagesList = []
for video in currentBatch:
resizedVideos = len(batchWithImagesList)
if not resizedVideos % 100:
print("Videos read and resized: {}/{}".format(resizedVideos, len(currentBatch)))
images = util.video_to_images(video[0])
images = self.resize_images_lrw(images)
batchWithImagesList.append((video[0], video[1], videoClassMap[video[1]], images))
batch_filename = os.path.join(dirpath, '{}_{}_batch_{}.{}'.format(datasetname, dataset_type, batchIndexer, "tfrecords"))
self.writeLRW(batch_filename, batchWithImagesList, dataset_type)
batch_stat_filename = os.path.join(dirpath, '{}_{}_batch_{}.{}'.format(datasetname, dataset_type, batchIndexer, "txt"))
self.write_written_images(batch_stat_filename, batchWithImagesList)
batchIndexer += 1
batchCounter += 1
return jsonify({
'brand': brand,
'batch': batch,
'tempRegion1': pred[0],
'tempRegion2': pred[1],
'time': time,
'version': '1',
'deviceStatus': 'deviceStatus'
})
@app.route('/api/load_model_config')
def api_load_model_config():
stage = request.args.get("stage")
if stage == 'produce':
return jsonify({'window_size': FEATURE_RANGE, 'block_size': int(FEATURE_RANGE / SPLIT_NUM)})
elif stage == 'transition':
return jsonify({'window_size': TRANSITION_FEATURE_RANGE, 'block_size': int(TRANSITION_FEATURE_RANGE / TRANSITION_SPLIT_NUM)})
else:
raise Exception('param error')
if __name__ == '__main__':
create_dir(MODEL_SAVE_DIR)
model_produce.load(MODEL_SAVE_DIR + load_best_model_prefix('produce'))
model_transition.load(MODEL_SAVE_DIR + load_best_model_prefix('transition'))
app.run(host='0.0.0.0')