specklegen 0.1.7

Generate sequences of random speckle patterns and defining optical flow fields.

Speckle Dataset Generation

Dataset Generation

We generate multi-frame synthetic speckle pattern image sequences and ground-truth flows that represent the underlying deformation of the sequence. Each sequence has a unique reference pattern and contains between 9,000 and 11,000 randomly generated ellipses of varying sizes, with major and minor axes ranging from 7 to 30 pixels. These ellipses are fully filled with random gray scale intensity gradients ranging from 0 to 255. We then backward warp each unique pattern with smooth and randomly generated spatial random deformation fields to generate deforming sequences. The random deformation fields are generated using GSTools, a library which uses covariance model to generate spatial random fields.

Sample Demo

Speckle Pattern	Flow Visualization

Run Speckle Generator

There are four arguments to be specified by the user. --output_path defines the directory where generated image sequences, ground-truth flows and flow visualizations will be saved. --seq_number and --seq_length represent the number of random speckle pattern sequences to generate and the number of frames per each sequence, respectively. Lastly, the --dimensions argument specifies the height and width of the output speckle patterns. Optionally, a user can specify --scales an argument that defines the max u and v flow magnitudes, default value is (10 10).

PyPI installation

This speckle dataset generator can be installed and used as follows:

Installation

conda create -n specklegen_env python=3.8
pip install specklegen==0.1.7

Usage

from specklegen.synthetic_data_generator import data_generator

# Define arguments
output_path = "./output" #output path
seq_number = 10 #number of sequences 
seq_length = 3 #number of frames per sequence
dimensions = (512, 512)  #output flow and sequence dimensions 
scales = (5, 7)  #max flow magnitudes of u and v fields, respectively

# Call function
data_generator(output_path, seq_number, seq_length, dimensions, scales)

Output Format

The output files include synthetic speckle pattern image sequences, .flo ground truth deformation fields containing the u and v components of the deformation, as well as flow visualization files (heatmaps of the u and v flows).

├── <output_path>/
│   ├── Sequences├──Seq1├──frame0001.png
│   │            │              .
│   │            │      ├──frame000n.png     
│   │            │ 
│   ├── Flow     ├──Seq1├──flow0001.flo
│   │            │              .
│   │            │      ├──frame000n-1.flo
│   │            │     
│   ├── Flow_vis ├──Seq1├──flow0001.png
│   │            │              .
│   │            │      ├──frame000n-1.png