Skip to main content

Invariant patch distances and clustering

Project description

This package includes methods for clustering patches and reconstructing images using invariant patch distances. The package is build upon the CUDA C++ project: invariant_patch_distances. The package has the following dependencies: an NVIDIA graphics card of compute capability 3.x or higher, CUDA 10.0 and CMake 3.10.

Example

The following example shows how to cluster patches and reconstruct an image using these clusters.

import patchdistances as pd

# Load image as Float32 numpy array. It can either be gray-scale or rgb.
img = load_image()

// Initialize the distance class.
// Alternatively: similarity_inv_dist.
aid = pd.affine_inv_dist()

// Greedy-k-center clustering of the image patches using the distance.
labels, indices_of_labels = aid.greedy_k_center(
    image=img,
    patch_shape=(7,7),
    clusters=20,
    first_patch_idx=0,
    solver=pd.solver_t.procrustes,
    func_family=pd.func_t.superlevelsets,
    use_imed=False,
    interpolation=pd.inter_t.bicubic,
)

// Reconstruct the image using the previously clustered patches.
// Alternatively: reconstruct_w_translation.
reconstructed_image, nearest_neighbor_labeling = aid.reconstruct(
    image=img,
    labels=labels,
    rec_type=pd.rec_t.Median,
    solver=pd.solver_t.procrustes,
    func_family=pd.func_t.superlevelsets,
    use_imed=False,
    interpolation=pd.inter_t.bicubic,
)

Project details


Release history Release notifications | RSS feed

This version

1.0

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

patchdistances-1.0.tar.gz (553.3 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page