event-lads 0.1

Locally Adaptive Decay Surfaces for event data representation.

Locally Adaptive Decay Surfaces (LADS)

Python package for event-based vision processing with LADS, paper preprint available here.

Installation

You can install the package using:

pip install event-lads

Usage

The LADS class is the core of this package and is used to integrate event data into locally adaptive decay surfaces. See the code snippet below for sample usage. A full implementation for all decay functions with a real event clip is provided in create_event_video.py.

import torch
import numpy as np
from event_lads import *
import matplotlib.pyplot as plt

# Initialize LADS
lads = LADS(
    H=64, W=64, # Spatial dims of input events and output surface
    device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu"),
    decay_func="er", # Options:["global-li", "er", "fft", "log"]
    reference_event_rate=0.1,
    decay_param=0.2,
    patch_size=8
)

# Generate dummy event data 
events = np.array([
    [0.1,  32, 32,  1], #(timestamp, x, y, polarity)
    [0.11, 10, 20,  1],
    [0.2,  40, 50, -1],
    [0.3,  32, 32,  1],
    [0.35, 32, 32,  1],
    [0.36, 10, 20,  1],
    [0.4,  40, 50, -1],
    [0.5,  32, 32,  1]
])


# Generate the surface by integrating the first block of events:
surface, patch_scores, patch_decay_factors = lads.integrateEvents(events[:4]) 
# patch_scores & patch_decay_factors are no longer needed but returned for analysis/visualisation.


frame = LADS_to_output_frame(surface, clip_val=3) # Convert the surface (tensor) to frame (ndarray).
plt.imshow(frame, cmap='gray')
plt.show()

# Update the surface by integrating more events:
surface, patch_scores, patch_decay_factors = lads.integrateEvents(events[4:]) 

frame = LADS_to_output_frame(surface, clip_val=3)
plt.imshow(frame, cmap='gray')
plt.show()