def test_pair_predict(pair_model_path, target_probe_path, target_gallery_path,
pid_path, score_path):
# todo
print(
"before load_model:----------------------------------------------------------------- "
)
model = load_model("./market_pair_pretrain.h5",
custom_objects={'triplet_hard_loss': triplet_hard_loss})
print(
"model finished:----------------------------------------------------------------- "
)
# model = ResNet50(weights='imagenet', include_top=False, input_tensor=Input(shape=(224, 224, 3)))
#model = Model(inputs=[model.layers[0].get_input_at(0)],
# outputs=[model.get_layer('ave_pooled_part').get_output_at(0)])
for layer in model.layers:
print("!!!!:", layer)
print("@@@@@", model.layers[-5])
#model = Model(inputs=[model.layers[0].get_input_at(0)],
# outputs=[model.layers[-4].get_output_at(0)] )
global_feature = model.layers[-5].get_output_at(0)
feature1 = GlobalAveragePooling2D(name='ave_pooled')(global_feature)
model = Model(inputs=[model.input], outputs=[feature1])
# model = Model(inputs=[model.input], outputs=[model.get_layer('avg_pool').output])
print(
"before test_predict:----------------------------------------------------------------- "
)
test_predict(model, target_probe_path, target_gallery_path, pid_path,
score_path)
def test_pair_predict(pair_model_path, target_probe_path, target_gallery_path, pid_path, score_path):
# todo
model = load_model(pair_model_path, custom_objects={'focal_loss_fixed': focal_loss_fixed})
# model = ResNet50(weights='imagenet', include_top=False, input_tensor=Input(shape=(224, 224, 3)))
model = Model(inputs=[model.get_layer('resnet50').get_input_at(0)],
outputs=[model.get_layer('resnet50').get_output_at(0)])
# model = Model(inputs=[model.input], outputs=[model.get_layer('avg_pool').output])
test_predict(model, target_probe_path, target_gallery_path, pid_path, score_path)
def test_rank_predict(rank_model_path, target_probe_path, target_gallery_path,
pid_path, score_path):
model = load_model(
rank_model_path,
custom_objects={'cross_entropy_loss': cross_entropy_loss})
model = Model(inputs=[model.get_layer('resnet50').get_input_at(0)],
outputs=[model.get_layer('resnet50').get_output_at(0)])
test_predict(model, target_probe_path, target_gallery_path, pid_path,
score_path)
def test_rank_predict(rank_model_path, target_probe_path, target_gallery_path, pid_path, score_path):
model = load_model(rank_model_path, custom_objects={'cross_entropy_loss': cross_entropy_loss})
for layer in model.get_layer('resnet50').layers:
if isinstance(layer, BatchNormalization):
print(layer.get_weights())
break
model = Model(inputs=[model.get_layer('resnet50').get_input_at(0)],
outputs=[model.get_layer('resnet50').get_output_at(0)])
test_predict(model, target_probe_path, target_gallery_path, pid_path, score_path)
def test_pair_predict(pair_model_path, target_probe_path, target_gallery_path,
pid_path, score_path):
model = load_model(pair_model_path)
model = Model(inputs=[model.get_layer('resnet50').get_input_at(0)],
outputs=[model.get_layer('resnet50').get_output_at(0)])
# model = Model(inputs=[model.input], outputs=[model.get_layer('avg_pool').output])
return test_predict(model, target_probe_path, target_gallery_path,
pid_path, score_path)