def build_index_and_search_through_it(images_features, file_index):
# Decide whether to do only image search or hybrid search
if not features_from_new_model_boolean:
# This is pure image search
image_index = vector_search.index_features(images_features, dims=4096)
search_key = get_index(input_image, file_index)
results = vector_search.search_index_by_key(search_key, image_index,
file_index)
print(results)
else:
word_vectors = vector_search.load_glove_vectors(glove_path)
# If we are searching for tags for an image
if not input_word:
# Work on a single image instead of indexing
search_key = get_index(input_image, file_index)
word_index, word_mapping = vector_search.build_word_index(
word_vectors)
results = vector_search.search_index_by_value(
images_features[search_key], word_index, word_mapping)
# If we are using words to search through our images
else:
image_index = vector_search.index_features(images_features,
dims=300)
results = vector_search.search_index_by_value(
word_vectors[input_word], image_index, file_index)
print(results)
def build_index_and_search_through_it(images_features, file_index):
# Decide whether to do only image search or hybrid search
if not features_from_new_model_boolean:
# This is pure image search
image_index = vector_search.index_features(images_features, dims=4096)
search_key = get_index(input_image, file_index)
results = vector_search.search_index_by_key(search_key, image_index, file_index)
print(results)
st.write('\n'.join('- `%s`' % elt for elt in results))
demo.show_top_n(9, results, search_by_img=False)
else:
word_vectors = vector_search.load_glove_vectors(glove_path)
# If we are searching for tags for an image
if not input_word:
# Work on a single image instead of indexing
search_key = get_index(input_image, file_index)
word_index, word_mapping = vector_search.build_word_index(word_vectors)
results = vector_search.search_index_by_value(images_features[search_key], word_index, word_mapping)
for result in results:
st.write(result)
image_index = vector_search.index_features(images_features, dims=300)
results = vector_search.search_index_by_value(word_vectors[result[1]], image_index, file_index)
st.write('\n'.join('- `%s`' % elt for elt in results))
demo.show_top_n(9, results, search_by_img=False)
# If we are using words to search through our images
else:
image_index = vector_search.index_features(images_features, dims=300)
st.write(word_vectors[input_word])
results = vector_search.search_index_by_value(word_vectors[input_word], image_index, file_index)
st.write('\n'.join('- `%s`' % elt for elt in results))
demo.show_top_n(9, results, search_by_img=False)
print(results)
def index_images(folder, features_path, mapping_path, model, features_from_new_model_boolean, glove_path):
print ("Now indexing images...")
# Use word vectors if leveraging the new model
if features_from_new_model_boolean:
word_vectors=vector_search.load_glove_vectors(glove_path)
else:
word_vectors=[]
# Use utiliy function
_, _, paths = load_paired_img_wrd(
folder=folder,
word_vectors=word_vectors,
use_word_vectors=features_from_new_model_boolean)
images_features, file_index = vector_search.generate_features(paths, model)
vector_search.save_features(features_path, images_features, mapping_path, file_index)
return images_features, file_index
def load_images_vectors_paths(glove_model_path, data_path):
word_vectors = vector_search.load_glove_vectors(glove_model_path)
images, vectors, image_paths = load_paired_img_wrd(data_path, word_vectors)
return images, vectors, image_paths, word_vectors
help='number of epochs to train on',
default=50)
return par
if __name__ == "__main__":
parser = build_parser()
options = parser.parse_args()
model_save_path = options.model_save_path
checkpoint_path = options.checkpoint_path
glove_path = options.glove_path
dataset_path = options.dataset_path
num_epochs = options.num_epochs
word_vectors = vector_search.load_glove_vectors(glove_path)
images, vectors, image_paths = load_paired_img_wrd(dataset_path,
word_vectors)
x, y = shuffle(images, vectors, random_state=2)
X_train, X_test, y_train, y_test = train_test_split(x,
y,
test_size=0.2,
random_state=2)
checkpointer = ModelCheckpoint(filepath=checkpoint_path,
verbose=1,
save_best_only=True)
custom_model = vector_search.setup_custon_model()
custom_model.fit(X_train,
y_train,
validation_data=(X_test, y_test),
st.header("Dataset")
st.subheader("Loading the data")
st.write("""
Let's start by loading our dataset, which consists of a total of a **1000 images**, divided in **20 classes**
with 50 images for each.
This dataset can be found [here](http://vision.cs.uiuc.edu/pascal-sentences/). *Credit to Cyrus
Rashtchian, Peter Young, Micah Hodosh, and Julia Hockenmaier.*
In addition, we load [GloVe](https://nlp.stanford.edu/projects/glove/) vectors pre-trained on Wikipedia,
which we will use when we incorporate text.
""")
with st.echo():
word_vectors = vector_search.load_glove_vectors("models/glove.6B")
images, vectors, image_paths = load_paired_img_wrd(
'dataset', word_vectors)
st.write("Here is what our load function looks like:")
show_source(load_paired_img_wrd)
all_labels = [fold.split("/")[1] for fold in image_paths]
st.write("Here is a list of our classes:",
'\n'.join('`%s`' % elt for elt in sorted(list(set(all_labels)))))
st.write(
"We now have a tensor of images of size %s, of word vectors of size %s, and a list of corresponding file "
"paths." % (images.shape, vectors.shape))
st.subheader("Visualizing the data")
st.write(
"Let's see what our data looks like, here is one example image from each class"