def generate_caption(indIm,
names,
model,
image_model,
caption_representations,
texts,
n=10):
# generate image representation for new image
image_filename = dataset_path + names[indIm]
print(names[indIm])
y_true = indicesByID(indIm, names)
features = extract_features_file(model, image_filename)
image_representation = image_model.predict(np.array([features]))
# compute score of all captions in the dataset
scores = np.dot(caption_representations, image_representation.T).flatten()
# compute indices of n best captions
indices = np.argpartition(scores, -n)[-n:]
indices = indices[np.argsort(scores[indices])]
# display them
y_pred = []
for i in [int(x) for x in reversed(indices)]:
print(i, texts[i])
y_pred.append(i)
# print('y_pred : ', y_pred)
print('y_true : ', y_true)
print('y_pred : ', y_pred)
y_pred = np.array(y_pred)
ranking_metrics = BinaryMetrics(y_true, y_pred)
print('k = ', n)
print('precision = ', ranking_metrics.precision_at(n))
print('recall = ', ranking_metrics.recall_at(n))
print('r precision = ', ranking_metrics.r_precision())
print('_______________________________')
def stats_generate_caption(names,
model,
image_model,
caption_representations,
texts,
n=500):
prec = []
recall = []
r_prec = []
for indIm in range(0, 100, 5):
thisPrec = []
thisRecall = []
thisRPrec = []
image_filename = dataset_path + names[indIm]
print(names[indIm])
y_true = indicesByID(indIm, names)
features = extract_features_file(model, image_filename)
image_representation = image_model.predict(np.array([features]))
scores = np.dot(caption_representations,
image_representation.T).flatten()
indices = np.argpartition(scores, -n)[-n:]
indices = indices[np.argsort(scores[indices])]
y_pred = []
for i in [int(x) for x in reversed(indices)]:
# print(i, texts[i])
y_pred.append(i)
y_pred = np.array(y_pred)
ranking_metrics = BinaryMetrics(y_true, y_pred)
thisRange = range(5, 500, 10)
for k in thisRange:
thisPrec.append(ranking_metrics.precision_at(k))
thisRecall.append(ranking_metrics.recall_at(k))
prec.append(thisPrec)
recall.append(thisRecall)
prec = pd.DataFrame(prec, columns=thisRange)
recall = pd.DataFrame(recall, columns=thisRange)
prec_moy = prec.mean()
prec_moy = prec_moy
recall_moy = recall.mean()
df = pd.DataFrame({
'Nbr de prédictions': thisRange,
'recall_moyen': recall_moy,
'prec_moy': prec_moy
})
plt.plot('Nbr de prédictions',
'recall_moyen',
data=df,
color='skyblue',
linewidth=2)
plt.plot('Nbr de prédictions',
'prec_moy',
data=df,
color='olive',
linewidth=2)
plt.legend()
plt.savefig('drive/MyDrive/benoit/figures/recall_prec')
def searchImagesFromImage(image_filename, model, image_model, image_representations, n=10) :
thisFeatures = extract_features_file(model, image_filename)
image_representation = image_model.predict(np.array([thisFeatures]))
scores = np.dot(image_representations,image_representation.T).flatten()
indices = np.argpartition(scores, -n)[-n:]
indices = indices[np.argsort(scores[indices])]
for i in [int(x) for x in reversed(indices)]:
print(scores[i], images[i])
cetteImage=images[i]
cetteImage=cetteImage[:-2]
def extract_features_dir(model, directory):
# extract features from each photo
all_features = []
for name in listdir(directory):
# load an image from file
filename = path.join(directory, name)
features = extract_features_file(model, filename)
features = np.squeeze(features)
all_features.append(features)
print('>%s' % name)
return all_features
def generate_caption(indIm, names, model, image_model, caption_representations, texts, n=10):
# generate image representation for new image
image_filename = dataset_path + names[indIm]
print(names[indIm])
y_true=indicesByID(indIm, names)
features = extract_features_file(model, image_filename)
image_representation = image_model.predict(np.array([features]))
# compute score of all captions in the dataset
scores = np.dot(caption_representations,image_representation.T).flatten()
# compute indices of n best captions
indices = np.argpartition(scores, -n)[-n:]
indices = indices[np.argsort(scores[indices])]
# display them
y_pred=[]
for i in [int(x) for x in reversed(indices)]:
print(i, texts[i])
def searchTextFromImageAndText(image_filename,
model,
image_model,
vocab,
caption,
caption_model,
ImageAndTextRepresentation,
texts,
features,
n=30):
features = extract_features_file(model, image_filename)
imageEncoded = image_model.predict(np.array([features]))
textEncoded = caption_model.predict(preprocess_texts([caption], vocab))
imageAndTextEncoded = np.concatenate((imageEncoded[0], textEncoded[0]))
scores = np.dot(ImageAndTextRepresentation,
imageAndTextEncoded.T).flatten()
indices = np.argpartition(scores, -n)[-n:]
indices = indices[np.argsort(scores[indices])]
for i in [int(x) for x in reversed(indices)]:
print(scores[i], texts[i])