python

Python API of LCD Toolbox

This is pybind version of Loop Closure Detection C++ libraries. Since python is a very powerful language for deep learning, this implementation can bind Loop Clousre Detection with Deep Learning Global descriptors.

What's inside?

We offer two main modules: dbow and vlad. Since two libraries have almost identical API, we will only explain with dbow here.

loopclosuretoolbox.dbow

For DBoW3, We offer (almost) the same API as C++. Here's an simple example.

Create vocabulary

We can either load vocabulary from path or create with HKMeans.

import loopclosuretoolbox.dbow as dbow

# create Vocabulary instance from file path
voc_load = dbow.Vocabulary("./config/sthereo_01_rgb_4_3.yaml")

# or create Vocabulary later
voc_created = dbow.Vocabulary(4,3) # k=4, l=3
training_features = ... # list of local descriptors to be used for train
voc_created.create(training_features)

managing database

Users can use add and query API just like C++.

import loopclosuretoolbox.dbow as dbow

# create Database with Vocabulary instance (voc)
db = dbow.Database(voc, False, 0)

# add entries to Database
for image in images:
    descs = get_descriptors(image) # user's implementation
    db.add(descs)


for query_image in query_images:
    q_descs = get_descriptors(query_image) # user's implementation
    results = db.query(q_descs, 5) # retrieve best 5 results

    # result is a tuple of (entry_id, score)
    for result in results:
        print(f'idx: {result[0]}, score: {result[1]}')

pairwise score

This one is new features (originally not included in DBoW3 libraries). For evaluation or visualization, users can get pairwise score between entries. Value lies between [0,1], and all values are double. Computations are parallelized for speed!

pscore = db.compute_pairwise_score()
cv2.imshow('pscore', pscore)
cv2.waitKey(0)