**params,
type_gen='train')
validation_generator = DataGeneratorBKB(test_keys,
labels,
**params,
type_gen='test')
# # Design model
if model_type == 'Conv3D':
model = create_model_Conv3D(dim,
n_sequence,
n_channels,
n_output,
set_pretrain=True)
else:
model = create_model_pretrain(dim, n_sequence, n_channels, n_output, 1.0)
load_model = True
start_epoch = 0
if load_model:
# weights_path = 'pretrain/mobileNetV2-BKB-3ds-48-0.55.hdf5'
# weights_path = 'BUPT-Conv3D-dataset02-transfer-0-0-0.hdf5' #'KARD-aug-RGBdif-01-0.13-0.17.hdf5'
weights_path = 'KARD-Conv3D-RGBdiff-crop-224-650-0.75-0.75.hdf5' #'BUPT-Conv3D-KARD-transfer-0-0-0.hdf5'
start_epoch = 650
model.load_weights(weights_path)
## Set callback
validate_freq = 10
# filepath= detail_weight+"-{epoch:02d}-{accuracy:.2f}-{val_accuracy:.2f}.hdf5"
# checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=False, period=validate_freq)
filepath = detail_weight + "-{epoch:02d}-{acc:.2f}-{val_acc:.2f}.hdf5"
index_sampling = get_sampling_frame(length_file) # get index to sampling
for j, n_pic in enumerate(index_sampling):
print(j, n_pic)
cap.set(cv2.CAP_PROP_POS_FRAMES, n_pic)
ret, frame = cap.read()
new_image = cv2.resize(frame, dim)
new_image = new_image / 255.0
X[0, j, :, :, :] = new_image
# cv2.imshow('Frame',frame)
# cv2.waitKey(500)
cap.release()
print(X.shape)
## Predict
weights_path = 'BUPT-augment-RGBdiff-120-0.90-0.91.hdf5'
# weights_path = 'KARD-aug-RGBdif-40-0.92-0.98.hdf5'
model = create_model_pretrain(dim, n_sequence, n_channels, n_output,
'MobileNetV2')
model.load_weights(weights_path)
X[0, ] = calculateRGBdiff(X[0, ])
for i in range(n_sequence):
cv2.imshow('Frame', X[0, i])
cv2.waitKey(500)
result = model.predict(X)
print(result)
# class_label = ['run','sit','stand','standup','walk']
kernel_size=(3, 3, 3),
activation='relu',
kernel_initializer='he_uniform'))
model.add(MaxPooling3D(pool_size=(2, 2, 2)))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dropout(.4))
model.add(Dense(24, activation='relu'))
model.add(Dropout(.4))
model.add(Dense(n_output, activation='softmax'))
model.compile(optimizer='sgd',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
return model
dim_1 = (120, 120)
dim_2 = (120, 120)
n_sequence_1 = 10
n_sequence_2 = 8
n_channels = 3
n_output = 5
model_1 = create_model_Conv3D(dim_1, n_sequence_1, n_channels, n_output)
model_2 = create_model_pretrain(dim_2, n_sequence_2, n_channels, n_output, 1.0)
model_3 = create_model_pretrain(dim_2, n_sequence_2, n_channels, n_output,
0.35)
print(model_1.summary())
print(model_2.summary())
print(model_3.summary())