cine-handler

cine-handler is the PyPI distribution for the cine_reader Python package, which reads Vision Research Phantom .cine files on macOS, Linux, and Windows.

The PyPI distribution contains only the Python package. The repository also includes a separate MATLAB implementation that stays in GitHub:

• Repository: https://github.com/rverleur/cine-reader
• MATLAB implementation: https://github.com/rverleur/cine-reader/tree/main/Matlab
• Python examples: https://github.com/rverleur/cine-reader/tree/main/python/examples
• Python API notes: https://github.com/rverleur/cine-reader/blob/main/python/API.md

Installation

Install from PyPI:

pip install cine-handler

Development install from GitHub:

pip install "git+https://github.com/rverleur/cine-reader.git"

Editable install from a local checkout:

pip install -e .[dev]

What The Package Provides

• High-level Cine reader for file metadata, frame access, frame iteration, trimming, and simple processing.
• Packed 10-bit decode with bundled native helpers and a NumPy fallback.
• Dead-pixel repair helpers for mono, Bayer/CFA, and RGB image arrays.
• Bilinear Bayer demosaic for raw color sensor frames.
• Low-memory frame statistics helpers for averaging and robust background estimation.

Quick Start

from pathlib import Path

from cine_reader import Cine

cine_path = Path("my_capture.cine")

with Cine(cine_path) as cine:
    print("frames:", cine.total_frames)
    print("range:", cine.first_frame_number, cine.last_frame_number)
    print("frame rate:", cine.frame_rate)
    print("exposure (s):", cine.exposure_time_seconds)

    cine.load_frame(cine.first_frame_number)
    raw_frame = cine.frame

    rgb_frame = cine.get_frame_rgb()
    average = cine.average_frames(cine.first_frame_number, cine.first_frame_number + 9)
    background = cine.mode_frames(cine.first_frame_number, cine.first_frame_number + 19)

Frame And Color Behavior

• Mono cines load as 2D arrays.
• Raw color CFA/Bayer cines load as 2D sensor mosaics by default.
• Interpolated 24-bit and 48-bit color payloads load as 3-channel arrays.
• Set debayer=True on Cine(...) or call debayer_frame() to convert raw CFA frames to RGB.
• Set remove_dead_pixels=True on Cine(...) or call replace_dead_pixels() to repair dead pixels.
• red_pixels, green_pixels, and blue_pixels expose raw CFA sensor samples for Bayer frames, using NaN at non-matching locations.

The Cine Class

Constructor

Cine(
    filename,
    keep_annotations=True,
    remove_dead_pixels=False,
    debayer=False,
    dead_value=None,
    dead_is_threshold=True,
    bayer_pattern="auto",
)

• filename: path to a .cine file.
• keep_annotations: keep annotation block bytes in annotation_data and annotation.
• remove_dead_pixels: repair dead pixels on each load_frame.
• debayer: debayer raw CFA/Bayer frames on each load_frame.
• dead_value: optional dead-pixel marker or threshold. If omitted, the value is inferred from camera metadata.
• dead_is_threshold: when True, values greater than or equal to dead_value are treated as dead pixels.
• bayer_pattern: "auto", "RGGB", "BGGR", "GRBG", or "GBRG".

Reader Lifecycle

• open_cine_file(filename): reopen a .cine file and parse the top-level metadata blocks.
• close_file(): close the active file handle.
• Context manager support: with Cine(path) as cine: ...

Metadata And Aliases

These properties are the most useful entry points for file metadata:

• first_frame_number
• total_frames
• last_frame_number
• frame_rate
• exposure_time_ns
• exposure_time_seconds
• exposure_time
• recording_datetime
• recording_date
• cfa_code
• bayer_pattern

The specification-aligned metadata objects are also available directly:

• file_header as CineHeader
• image_header as BitmapHeader
• camera_setup as Setup
• image_locations as ImageOffsets

Frame Access

• load_frame(image_no, convert_bgr_to_rgb=False)
  • Loads a single frame by global frame number.
  • convert_bgr_to_rgb=True converts 3-channel payloads to RGB on load.
• next_frame(increment=1, convert_bgr_to_rgb=False)
  • Loads the next frame relative to current_frame.
  • Negative increments are allowed.
• frame and image
  • Aliases for the currently loaded NumPy array.
• get_frame_rgb(image_no=None, bayer_pattern="auto")
  • Returns the current frame, or a selected frame, as RGB.
  • Raw CFA/Bayer frames are demosaiced first.

Example:

with Cine("capture.cine") as cine:
    cine.load_frame(cine.first_frame_number)
    raw = cine.frame
    rgb = cine.get_frame_rgb()

Pixel Repair And Demosaic

• replace_dead_pixels(dead_value=None, dead_is_threshold=True)
  • Repairs dead pixels on the current frame.
  • Works for mono, raw Bayer/CFA, and RGB data.
• debayer_frame(bayer_pattern="auto")
  • Converts the current raw CFA/Bayer frame to RGB in place.

Example:

with Cine("capture.cine") as cine:
    cine.load_frame(cine.first_frame_number)
    cine.replace_dead_pixels()
    cine.debayer_frame()
    rgb = cine.frame

Multi-Frame Operations

• average_frames(start_frame, end_frame, replace_dead_pixels=False, chunk_size=8)
  • Computes a per-pixel average over an inclusive frame range.
• mode_frames(start_frame, end_frame, replace_dead_pixels=False, method="auto", q_bg=0.80, k_sigma=2.5, min_keep=3, max_keep=96, stack_limit=128)
  • Computes a robust bright-background estimate.
  • method="mad" uses a full-stack quantile/MAD estimator.
  • method="topk" uses a bounded-memory top-k estimator.
  • method="auto" chooses between them from the range length.
• load_frames_batch(start_frame, count)
  • Loads consecutive frames into one stacked array.
  • Output shape is [H, W, N] for mono/raw CFA and [H, W, 3, N] for color.
• save_frames_to_new_file(output_filename, start_frame, end_frame)
  • Writes a trimmed .cine file for an inclusive frame range.

Example:

with Cine("capture.cine") as cine:
    first = cine.first_frame_number
    last = min(first + 49, cine.last_frame_number)

    avg = cine.average_frames(first, last, chunk_size=8)
    bg = cine.mode_frames(first, last, method="topk")
    batch = cine.load_frames_batch(first, 5)
    cine.save_frames_to_new_file("trimmed.cine", first, last)

Standalone Helper Functions

Packed 10-bit Decode

• unpack_10bit_data(data)
  • Decodes Phantom packed 10-bit payloads.
  • Uses a bundled native unpack helper when one is available.
  • Falls back to NumPy automatically.
• unpack_10bit_numpy(data)
  • Pure-NumPy packed 10-bit decoder.

To force the NumPy path:

export CINE_READER_DISABLE_C_UNPACK=1

Dead-Pixel Repair

• replace_dead_pixels(frame, dead_value=4095, dead_is_threshold=False, bayer_raw=False)
  • Dispatches to the right repair strategy for mono, Bayer, or RGB input.
• replace_dead_pixels_mono(frame, dead_value=4095, dead_is_threshold=False)
  • Repairs dead pixels in a 2D mono frame from valid 8-neighbor values.
• replace_dead_pixels_bayer(frame, dead_value=4095, dead_is_threshold=False)
  • Repairs a 2D raw Bayer/CFA frame phase-by-phase before demosaic.
• replace_dead_pixels_rgb(frame, dead_value=4095, dead_is_threshold=False)
  • Repairs each color channel independently in a 3D image.

Example:

from cine_reader import replace_dead_pixels_bayer

clean = replace_dead_pixels_bayer(raw_bayer_frame, dead_value=4095)

Demosaic

• demosaic_bilinear(frame, pattern="RGGB")
  • Bilinear Bayer demosaic for raw CFA frames.
  • Supported patterns are "RGGB", "BGGR", "GRBG", and "GBRG".

Example:

from cine_reader import demosaic_bilinear

rgb = demosaic_bilinear(raw_bayer_frame, pattern="RGGB")

Frame Statistics

• average_from_frame_iter(frames, out_dtype=None, chunk_size=8)
  • Computes a per-pixel mean from an iterable of equally shaped frames.
  • Returns (average_frame, frame_count).
• robust_background_mad_stack(frames, q_bg=0.80, k_sigma=2.5, min_keep=3, out_dtype=np.uint16)
  • Full-stack bright-background estimator based on quantile and MAD rejection.
• robust_background_topk(frames, frame_count, q_bg=0.80, min_keep=3, max_keep=96, out_dtype=np.uint16)
  • Bounded-memory bright-background estimator that keeps the top-k samples per pixel.

Example:

from cine_reader import average_from_frame_iter, robust_background_topk

avg, count = average_from_frame_iter(frames, chunk_size=16)
bg = robust_background_topk(frames, frame_count=count, q_bg=0.80)

Metadata Classes

These classes are returned by the reader for direct access to CINE metadata:

• CineHeader
  • File header fields such as FirstImageNo, ImageCount, and OffImageOffsets.
• BitmapHeader
  • Image header fields such as biWidth, biHeight, and biBitCount.
• ImageOffsets
  • Frame byte-offset table as pImage.
• Setup
  • Camera setup metadata including frame rate, exposure, bit depth, color flags, and descriptive strings.

Example:

with Cine("capture.cine") as cine:
    print(cine.file_header.FirstImageNo)
    print(cine.image_header.biWidth, cine.image_header.biHeight)
    print(cine.camera_setup.FrameRate, cine.camera_setup.RealBPP)

Included Native Helpers

The wheel ships with runtime unpack libraries for packed 10-bit payloads:

• unpack_data_win32.dll
• unpack_data_win64.dll
• unpack_data_elf64.so
• unpack_data_arm64.dylib

If no matching helper can be loaded, cine-handler falls back to the NumPy decoder automatically.

MATLAB Implementation

The PyPI package is intentionally Python-only. If you also need the MATLAB reader, use the same GitHub repository:

• MATLAB package entry point: https://github.com/rverleur/cine-reader/blob/main/Matlab/Cine.m
• MATLAB setup and examples: https://github.com/rverleur/cine-reader/blob/main/Matlab/README.md

More Examples

• Basic usage: https://github.com/rverleur/cine-reader/blob/main/python/examples/basic_usage.py
• Trimming and background estimation: https://github.com/rverleur/cine-reader/blob/main/python/examples/trim_and_background.py
• Performance test: https://github.com/rverleur/cine-reader/blob/main/python/examples/performance_test.py