Skip to content

⚡ Bolt: Optimize TripletDataGenerator sampling to O(1) #13

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
google-labs-jules wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

⚡ Bolt: Optimize TripletDataGenerator sampling to O(1) #13

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 18 additions & 6 deletions deeptuner/datagenerators/triplet_data_generator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
from tensorflow.keras.preprocessing.image import ImageDataGenerator, load_img, img_to_array
from sklearn.preprocessing import LabelEncoder
import numpy as np
import random
from tensorflow.keras.applications import resnet50 as resnet

class TripletDataGenerator(tf.keras.utils.Sequence):
Expand All @@ -14,6 +15,14 @@ def __init__(self, image_paths, labels, batch_size, image_size, num_classes):
self.label_encoder = LabelEncoder()
self.encoded_labels = self.label_encoder.fit_transform(labels)
self.image_data_generator = ImageDataGenerator(preprocessing_function=resnet.preprocess_input)

# Precompute label to paths map for O(1) positive sampling
self.label_to_paths = {}
for path, label in zip(self.image_paths, self.encoded_labels):
if label not in self.label_to_paths:
self.label_to_paths[label] = []
self.label_to_paths[label].append(path)

self.on_epoch_end()
print(f"Initialized TripletDataGenerator with {len(self.image_paths)} images")

Expand All @@ -40,12 +49,15 @@ def _generate_triplet_batch(self, batch_image_paths, batch_labels):
anchor_path = batch_image_paths[i]
anchor_label = batch_labels[i]

positive_path = np.random.choice(
[p for p, l in zip(self.image_paths, self.encoded_labels) if l == anchor_label]
)
negative_path = np.random.choice(
[p for p, l in zip(self.image_paths, self.encoded_labels) if l != anchor_label]
)
# Optimized positive sampling: O(1) lookup
positive_path = random.choice(self.label_to_paths[anchor_label])

# Optimized negative sampling: Rejection sampling O(1) expected
while True:
idx = np.random.randint(0, len(self.image_paths))
if self.encoded_labels[idx] != anchor_label:
negative_path = self.image_paths[idx]
break

anchor_image = load_img(anchor_path, target_size=self.image_size)
positive_image = load_img(positive_path, target_size=self.image_size)
Expand Down