def _load_net(self):
# Load feature extractor
feature_extractor = feature_extractors.StridedInflatedEfficientNet()
checkpoint = engine.load_weights(self.feature_extractor_weights)
feature_extractor.load_state_dict(checkpoint)
feature_extractor.eval()
# Load a logistic regression classifier
gesture_classifier = LogisticRegression(
num_in=feature_extractor.feature_dim, num_out=30)
checkpoint = engine.load_weights(self.gesture_classifier_weights)
gesture_classifier.load_state_dict(checkpoint)
gesture_classifier.eval()
# Combine both models
self.net = Pipe(feature_extractor, gesture_classifier)
num_in=feature_extractor.feature_dim, num_out=81)
checkpoint = engine.load_weights(
'resources/fitness_activity_recognition/mobilenet_logistic_regression.ckpt'
)
gesture_classifier.load_state_dict(checkpoint)
gesture_classifier.eval()
# Load MET value converter
met_value_converter = calorie_estimation.METValueMLPConverter()
checkpoint = torch.load(
'resources/calorie_estimation/mobilenet_features_met_converter.ckpt')
met_value_converter.load_state_dict(checkpoint)
met_value_converter.eval()
# Concatenate feature extractor with downstream nets
net = Pipe(feature_extractor,
feature_converter=[gesture_classifier, met_value_converter])
post_processors = [
PostprocessClassificationOutput(INT2LAB, smoothing=8, indices=[0]),
calorie_estimation.CalorieAccumulator(weight=weight,
height=height,
age=age,
gender=gender,
smoothing=12,
indices=[1])
]
display_ops = [
realtimenet.display.DisplayFPS(expected_camera_fps=net.fps,
expected_inference_fps=net.fps /
net.step_size),
# Load feature extractor
feature_extractor = feature_extractors.StridedInflatedMobileNetV2()
checkpoint = engine.load_weights(
'resources/backbone/strided_inflated_mobilenet.ckpt')
feature_extractor.load_state_dict(checkpoint)
feature_extractor.eval()
# Load MET value converter
met_value_converter = calorie_estimation.METValueMLPConverter()
checkpoint = engine.load_weights(
'resources/calorie_estimation/mobilenet_features_met_converter.ckpt')
met_value_converter.load_state_dict(checkpoint)
met_value_converter.eval()
# Concatenate feature extractor and met converter
net = Pipe(feature_extractor, met_value_converter)
post_processors = [
calorie_estimation.CalorieAccumulator(weight=weight,
height=height,
age=age,
gender=gender,
smoothing=12)
]
display_ops = [
realtimenet.display.DisplayFPS(expected_camera_fps=net.fps,
expected_inference_fps=net.fps /
net.step_size),
realtimenet.display.DisplayDetailedMETandCalories(),
]
# Load feature extractor
feature_extractor = feature_extractors.StridedInflatedEfficientNet()
checkpoint = engine.load_weights('resources/backbone/strided_inflated_efficientnet.ckpt')
feature_extractor.load_state_dict(checkpoint)
feature_extractor.eval()
# Load a logistic regression classifier
gesture_classifier = LogisticRegression(num_in=feature_extractor.feature_dim,
num_out=30)
checkpoint = engine.load_weights('resources/gesture_detection/efficientnet_logistic_regression.ckpt')
gesture_classifier.load_state_dict(checkpoint)
gesture_classifier.eval()
# Concatenate feature extractor and met converter
net = Pipe(feature_extractor, gesture_classifier)
# Create inference engine, video streaming and display instances
inference_engine = engine.InferenceEngine(net, use_gpu=use_gpu)
video_source = camera.VideoSource(camera_id=camera_id,
size=inference_engine.expected_frame_size,
filename=path_in)
video_stream = camera.VideoStream(video_source,
inference_engine.fps)
postprocessor = [
PostprocessClassificationOutput(INT2LAB, smoothing=4)
]
label2int,
num_timesteps=num_timesteps)
valid_loader = generate_data_loader(os.path.join(
path_in,
f"features_valid_num_layers_to_finetune={num_layers_to_finetune}"),
label_names,
label2int,
num_timesteps=None,
batch_size=1,
shuffle=False)
# modeify the network to generate the training network on top of the features
gesture_classifier = LogisticRegression(
num_in=feature_extractor.feature_dim, num_out=len(label_names))
if num_layers_to_finetune > 0:
net = Pipe(custom_classifier_bottom, gesture_classifier)
else:
net = gesture_classifier
net.train()
if use_gpu:
net = net.cuda()
lr_schedule = {0: 0.0001, 40: 0.00001}
num_epochs = 60
best_model_state_dict = training_loops(net, train_loader, valid_loader,
use_gpu, num_epochs, lr_schedule)
# Save best model
if isinstance(net, Pipe):
best_model_state_dict = {