desc_img = descriptor_build(desc, x_img)
desc_warped_img = descriptor_build(desc, x_warped_img)
concat = tf.keras.layers.Concatenate()([det_img['logits'], det_warped_img['logits'], desc_img, desc_warped_img])
model = tf.keras.Model([img_in, warped_img_in], concat)
model.summary()
"""
tf.keras.utils.plot_model(model, config['model_visual'], show_shapes=True)
"""
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
model.compile(optimizer=tf.keras.optimizers.Adam(lr=config['learning_rate']), loss=total_loss,
metrics=[precision_metric(det_img['pred']), recall_metric(det_img['pred']),
warped_precision_metric(det_warped_img['pred']), warped_recall_metric(det_warped_img['pred']),
threshold_precision_metric(det_img['pred']), threshold_recall_metric(det_img['pred']),
warped_threshold_precision_metric(det_warped_img['pred']), warped_threshold_recall_metric(det_warped_img['pred']),
repeatability_metric(det_img['pred'], det_warped_img['pred'])],
options=run_options, run_metadata=run_metadata)
if not config['pretrained_model']:
pass
else:
model.load_weights(basepath + '/' + config['pretrained_weights'], by_name=True)
"""
model._make_train_function()
with open(basepath + '/' + config['pretrained_optimizer'], 'rb') as opt:
weight_values = pickle.load(opt)
model.optimizer.set_weights(weight_values)
sess = tf.Session(config=config)
tf.keras.backend.set_session(sess)
basepath = '/home/ubuntu/data'
with open('configs/config_sp_hpatches_descriptors.yaml', 'r') as f:
config = yaml.load(f)
model = tf.keras.models.load_model(basepath + '/' + config['model'],
custom_objects={
'total_loss':
total_loss,
'precision':
precision_metric(0),
'recall':
recall_metric(0),
'warped_precision':
warped_precision_metric(0),
'warped_recall':
warped_recall_metric(0),
'threshold_precision':
threshold_precision_metric(0),
'threshold_recall':
threshold_recall_metric(0),
'warped_threshold_precision':
warped_threshold_precision_metric(0),
'warped_threshold_recall':
warped_threshold_recall_metric(0),
'repeatability':
repeatability_metric(
np.zeros((1, 1), np.int32),
d = detector_head()
det = detector_build(d, encoder_layers, **config)
model = tf.keras.Model(img_in, det['logits'])
model.summary()
"""
tf.keras.utils.plot_model(model, config['model_visual'], show_shapes=True)
"""
run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
model.compile(optimizer=tf.keras.optimizers.Adam(lr=config['learning_rate']),
loss=detector_loss,
metrics=[
precision_metric(det['pred']),
recall_metric(det['pred']),
threshold_precision_metric(det['pred']),
threshold_recall_metric(det['pred'])
],
options=run_options,
run_metadata=run_metadata)
if not config['pretrained_model']:
pass
else:
model.load_weights(basepath + '/' + config['pretrained_weights'])
"""
model._make_train_function()
with open(basepath + '/' + config['pretrained_optimizer'], 'rb') as opt:
weight_values = pickle.load(opt)
model.optimizer.set_weights(weight_values)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
tf.keras.backend.set_session(sess)
d2l = lambda d: [dict(zip(d, e)) for e in zip(*d.values())]
basepath = '/home/ubuntu/data'
with open('configs/config_mp_coco_export.yaml', 'r') as f:
config = yaml.load(f)
model = tf.keras.models.load_model(basepath + '/' + config['model'],
custom_objects={'detector_loss': detector_loss,
'precision': precision_metric(0), 'recall': recall_metric(0),
'threshold_precision': threshold_precision_metric(0),
'threshold_recall': threshold_recall_metric(0)})
model.summary()
picklefile = Path(basepath, config['picklefile'])
with open(picklefile, 'rb') as handle:
files = pickle.load(handle)
output_dir = Path(basepath, config['export_name'])
if not output_dir.exists():
os.makedirs(output_dir, exist_ok=True)
with open(basepath + '/' + config['export_name'] + '/' + 'config.yml', 'w') as f:
yaml.dump(config, f, default_flow_style=False)