SEGFAST

Installation • Benchmarks • Getting Started

---

segfast is a library for interacting with SEG-Y seismic data. Main features are:

• Faster access to read data: both traces headers and values
• Optional bufferization, where user can provide a preallocated memory to load the data into
• Convenient API that relies on numpy.memmap for most operations, while providing segyio as a fallback engine

Installation

# pip / pip3
pip3 install segfast

# developer version (add `--depth 1` if needed)
git clone https://github.com/analysiscenter/segfast.git

Benchmarks

Timings for reading data along various projections:

	slide_i	slide_x	slide_d	crop (256, 256, 500)	batch (20, 256, 256, 500)
segyio	2.58254	7.16672	3041.3	941.285	16104.4
segfast	1.48056	3.37418	50.1355	82.0574	2761.94
segfast segyio engine	2.92379	5.69101	225.13	117.571	3968.81
seismiqb	1.46763	3.45154	50.3333	151.877	2738.86
seismiqb+HDF5	1.04213	1.93414	1.80567	81.3581	2616.83
segfast quantized	0.252452	0.518485	56.6672	7.71151	1212.74

Getting started

After installation just import segfast into your code. A quick demo of our primitives and methods:

import segfast

# Open file and read some meta info. Engine can be `segyio` or `memmap`
segfast_file = segfast.open('/path/to/cube.sgy', engine='memmap')

# Load requested headers as dataframe
segfast_file.load_headers(['INLINE_3D', 'CROSSLINE_3D', ...])

# Data access. All methods support optional buffer as target memory
segfast_file.load_traces([123, 333, 777], buffer=None)
segfast_file.load_depth_slices([5, 10, 15], buffer=None)

# Convert data format to IEEE float32: speeds up operations by a lot
segfast_file.convert(format=5)

You can get more familiar with the library, its functional and timings by reading examples.