pixtreme 0.8.4

pixtreme: A High-Performance Graphics Library with CUDA Support

🚀 pixtreme

Blazing-fast GPU-accelerated image processing for Python

🌟 Highlights

• ⚡ Lightning Fast: CUDA-optimized kernels deliver real-time performance
• 🎨 Professional Color Pipeline: Full ACES workflow, 3D LUTs, 10-bit precision
• 🧠 AI-Ready: Seamless integration with ONNX, PyTorch, and TensorRT
• 🔗 Zero-Copy Interop: DLPack support for PyTorch, TensorFlow, JAX
• 📊 Extensive Format Support: OpenEXR, JPEG, PNG, TIFF, and more

📋 Table of Contents

• Breaking Changes
• Features
• Installation
• Quick Start
• API Reference
• Performance Notes
• License

⚠️ Breaking Changes

v0.8.4: Critical Bugfix for Type Hints (2025-10-27)

🔥 CRITICAL BUGFIX: Fixed runtime import errors in non-PyTorch environments - v0.8.3 still had issues with type hint evaluation causing AttributeError when PyTorch is not installed.

What was fixed:

• Type hints wrapped in string literals: All torch.device and torch.Tensor references in type annotations now use string literals ("torch.Tensor") to prevent runtime evaluation
• Root cause: While v0.8.3 added from __future__ import annotations, some environments (Python 3.13, pydantic) still evaluate annotations at runtime
• Impact: pixtreme-core now imports successfully in all environments, regardless of PyTorch installation status

Who should upgrade:

• All v0.8.3 users immediately - v0.8.3 is broken in non-PyTorch environments
• Users running Python 3.13 or using libraries that evaluate annotations at runtime

Upgrade:

pip install --upgrade pixtreme-core>=0.8.4
# or
pip install --upgrade pixtreme[all]>=0.8.4

Technical details: Changed type hints from torch.device to "torch.device" in dlpack.py:25,26,47,64 to ensure compatibility with all annotation evaluation strategies.

v0.8.3: nvimgcodec v0.6.0+ Required (2025-10-27)

⚠️ BREAKING CHANGE: nvimgcodec >= 0.6.0 now required - v0.8.3 drops support for nvimgcodec v0.5.x to simplify code and adopt the latest API.

What changed:

• Dependency updated: nvidia-nvimgcodec-cu12[all]>=0.6.0 (was >=0.5.0)
• dlpack.py: Improved TYPE_CHECKING pattern for torch imports (fixes community-reported issue)
• imread.py: Simplified to use nvimgcodec v0.6.0 API only (removed backward compatibility code)

nvimgcodec v0.6.0 API Changes:

• DecodeSource class removed from Python API (deprecated in v0.6.0-beta.6)
• Direct file path passing to Decoder.read() is the new standard
• Cleaner, simpler API with better performance

Who should upgrade:

• All users - v0.8.3 requires nvimgcodec >= 0.6.0
• If you need nvimgcodec < 0.6.0, stay on pixtreme v0.8.2

Upgrade:

pip install --upgrade pixtreme-core>=0.8.3
# or
pip install --upgrade pixtreme[all]>=0.8.3

Note: This will automatically upgrade nvimgcodec to v0.6.0+ due to dependency requirements.

v0.8.2: Critical Bugfix (2025-10-27)

Fixed import error when PyTorch not installed - v0.8.0 had a critical bug where pixtreme-core would fail to import in environments without PyTorch.

What was fixed:

• Added from __future__ import annotations to prevent runtime evaluation of type hints
• torch.device type annotations no longer cause AttributeError when torch is not installed
• Module imports now succeed with TORCH_AVAILABLE=False flag set correctly

Who should upgrade:

• All v0.8.0 users - v0.8.0 is broken in non-PyTorch environments
• Users who install pixtreme-core without the full pixtreme[all] bundle

Upgrade:

pip install --upgrade pixtreme-core>=0.8.2
# or
pip install --upgrade pixtreme[all]>=0.8.2

v0.8.0: Morphology Operations Reorganization

Morphology module moved from core to filter - The erode function and related morphology operations have been relocated to the pixtreme-filter package for better organization.

API Changes:

• erode() moved from pixtreme-core to pixtreme-filter
• New operations added: dilate(), morphology_open(), morphology_close(), morphology_gradient()
• All morphology functions now in unified pixtreme_filter.morphology module

Migration: Update your imports:

# OLD (v0.7.x and earlier)
from pixtreme_core import erode

# NEW (v0.8.0+)
from pixtreme_filter.morphology import erode, dilate, morphology_open, morphology_close, morphology_gradient

# Or use the convenience import
import pixtreme as px
px.erode(image, ksize=5)  # Still works if pixtreme-filter is installed

Installation: Ensure pixtreme-filter is installed:

pip install pixtreme[filter]  # or pixtreme[all] for all features

v0.7.3: Type System & Build Modernization

Python 3.12+ now required - pixtreme v0.7.3 drops Python 3.10/3.11 support and requires Python 3.12 or later.

Type System Improvements:

• Full mypy compatibility with strict type checking
• Improved error messages with detailed value reporting
• 15+ type annotation bugs fixed across core modules

Build System Optimization:

• License classifier added to all packages (MIT)
• Issues URL standardized across packages
• Pre-commit hooks for local quality checks (mypy, ruff, version consistency)
• sdist/wheel metadata improvements

Developer Experience:

• uv-native pre-commit config (no virtualenv overhead)
• Better error reporting in dtype conversions and validation
• Comprehensive metadata for PyPI display

Migration: Update Python to 3.12+ and reinstall:

# Ensure Python 3.12 or later
python --version  # Should show 3.12.x or 3.13.x

pip install --upgrade pixtreme>=0.7.3

v0.6.3: Bug Fixes and API Restoration

v0.6.3 restores missing APIs from v0.6.0 and includes important bug fixes:

Restored APIs (accidentally removed in v0.6.0):

• I/O functions: destroy_all_windows(), imdecode(), imencode(), waitkey()
• Type conversions: to_dtype(), to_float16(), to_float64()
• Transform functions: affine_transform(), get_inverse_matrix()

Bug Fixes:

• imwrite() now returns bool (success/failure) instead of None
• LUT parser improved to skip non-numeric lines in .cube files
• Comprehensive test suite added (297 tests, 99.7% pass rate)

Migration: If you encountered AttributeError for these functions in v0.6.0, upgrade to v0.6.3:

pip install --upgrade pixtreme>=0.6.3

v0.6.0: Modular Package Structure

pixtreme is now split into modular packages for better flexibility:

• pixtreme-core: Core functionality (always installed)
• pixtreme-aces: ACES color management (optional)
• pixtreme-filter: Image filtering (optional)
• pixtreme-draw: Drawing primitives (optional)
• pixtreme-upscale: Deep learning upscalers (optional)

Backward compatibility: import pixtreme as px still works with all installed packages.

Migration: No code changes needed. pip install pixtreme[all] for previous behavior.

v0.5.2: _cp Functions Removed

🚨 IMPORTANT: v0.5.1 is broken - use v0.5.2 or later

v0.5.1 contains a critical bug that prevents import. If you have v0.5.1 installed, upgrade immediately:

pip install --upgrade pixtreme>=0.5.2

_cp functions have been removed from the main pixtreme package in v0.5.2 (attempted in v0.5.1 but broken).

The following functions are now available only in the pixtreme-legacy package:

• apply_lut_cp → Use apply_lut instead
• uyvy422_to_ycbcr444_cp → Use uyvy422_to_ycbcr444 instead
• ndi_uyvy422_to_ycbcr444_cp → Use ndi_uyvy422_to_ycbcr444 instead
• yuv420p_to_ycbcr444_cp → Use yuv420p_to_ycbcr444 instead
• yuv422p10le_to_ycbcr444_cp → Use yuv422p10le_to_ycbcr444 instead

Migration Path

Option 1: Use pixtreme-legacy (Temporary):

pip install pixtreme-legacy

from pixtreme_legacy import apply_lut_cp
result = apply_lut_cp(image, lut)  # Works without warnings

Option 2: Migrate to standard functions (Recommended):

from pixtreme import apply_lut
result = apply_lut(image, lut)  # No warning

Legacy Support

If you need continued support for _cp functions, install pixtreme-legacy:

pip install pixtreme-legacy

✨ Features

🎯 Image Processing

• 11 Interpolation Methods: Nearest, Linear, Cubic, Area, Lanczos (2/3/4), Mitchell, B-Spline, Catmull-Rom
• Advanced Transforms: Affine transformations, tiling with overlap blending
• Morphological Operations: Erosion with custom kernels
• GPU-Accelerated Filters: Gaussian blur, custom convolutions

🎨 Color Science

• Color Spaces: BGR/RGB, HSV, YCbCr, YUV (4:2:0, 4:2:2), Grayscale
• ACES Pipeline: Complete Academy Color Encoding System workflow
• 3D LUT Processing: Trilinear and tetrahedral interpolation
• 10-bit Precision: Professional video color accuracy

🤖 Deep Learning

• Multi-Backend Support: ONNX Runtime, PyTorch, TensorRT
• Super Resolution: Built-in upscaling with various models
• Batch Processing: Efficient multi-image inference
• Model Optimization: Automatic conversion and optimization tools

🔧 Advanced Features

• Memory I/O: Encode/decode images in memory
• Hardware Acceleration: NVIDIA nvimgcodec support
• Drawing Tools: GPU-accelerated shapes and text rendering
• Framework Integration: Zero-copy tensor sharing via DLPack

🚀 Installation

Requirements

• Python >= 3.12
• CUDA Toolkit 12.x
• NVIDIA GPU with compute capability >= 6.0

Quick Install

v0.6.0 introduces modular packages - install only what you need:

# Core only (I/O, color, transform, utils)
pip install pixtreme

# Core + ACES color management
pip install pixtreme[aces]

# Core + image filters
pip install pixtreme[filter]

# Core + drawing primitives
pip install pixtreme[draw]

# Core + deep learning upscalers
pip install pixtreme[upscale]

# All features
pip install pixtreme[all]

# All features + legacy support + TensorRT
pip install pixtreme[full]

Individual Packages

You can also install packages individually:

pip install pixtreme-core      # Core functionality
pip install pixtreme-aces      # ACES color management
pip install pixtreme-filter    # Image filtering
pip install pixtreme-draw      # Drawing primitives
pip install pixtreme-upscale   # Deep learning upscalers
pip install pixtreme-legacy    # Legacy _cp functions

Development Setup

# Clone the repository
git clone https://github.com/sync-dev-org/pixtreme.git
cd pixtreme

# Install uv package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# Setup development environment
uv python pin 3.12
uv sync --extra dev --extra opencv

💡 Quick Start

import pixtreme as px

# Read image directly to GPU as float32 (BGR format)
image = px.imread("photo.jpg")

# All operations work on GPU memory
image_rgb = px.bgr_to_rgb(image)
image_hsv = px.rgb_to_hsv(image_rgb)

# High-quality resize with 11 interpolation methods
image = px.resize(image, (1920, 1080), interpolation=px.INTER_LANCZOS4)

# Choose backend based on your needs
upscaler = px.OnnxUpscaler("models/realesrgan.onnx")     # Balanced
# upscaler = px.TrtUpscaler("models/realesrgan.trt")     # Fastest
# upscaler = px.TorchUpscaler("models/realesrgan.pth")   # Most flexible

# Upscale with single method call
upscaled = upscaler.get(image)

# Professional color grading with 3D LUT
lut = px.read_lut("cinematic_look.cube")
graded = px.apply_lut(image, lut, interpolation=1)  # Tetrahedral

# Save with format-specific options
px.imwrite("output.jpg", graded, param=95)

📖 API Reference

🎮 Device Management

Function	Parameters	Returns	Description
get_device_id()	-	int	Get current CUDA device ID
get_device_count()	-	int	Get number of available GPUs
Device(id)	id: int	Context manager	Context manager for device selection

🎨 Color Module (pixtreme.color)

Basic Color Conversions

Function	Parameters	Returns	Description
bgr_to_rgb(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert BGR to RGB format
rgb_to_bgr(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert RGB to BGR format
bgr_to_grayscale(image)	image: cp.ndarray	cp.ndarray	Convert BGR to grayscale
rgb_to_grayscale(image)	image: cp.ndarray	cp.ndarray	Convert to grayscale (Rec.709)
bgr_to_hsv(image)	image: cp.ndarray	cp.ndarray	Convert BGR to HSV
hsv_to_bgr(image)	image: cp.ndarray	cp.ndarray	Convert HSV to BGR
rgb_to_hsv(image)	image: cp.ndarray	cp.ndarray	Convert RGB to HSV
hsv_to_rgb(image)	image: cp.ndarray	cp.ndarray	Convert HSV to RGB

YCbCr/YUV Conversions

Function	Parameters	Returns	Description
bgr_to_ycbcr(image)	image: cp.ndarray	cp.ndarray	BGR to YCbCr (10-bit precision)
rgb_to_ycbcr(image)	image: cp.ndarray	cp.ndarray	RGB to YCbCr (10-bit precision)
ycbcr_to_bgr(image)	image: cp.ndarray	cp.ndarray	YCbCr to BGR conversion
ycbcr_to_rgb(image)	image: cp.ndarray	cp.ndarray	YCbCr to RGB conversion
ycbcr_to_grayscale(image)	image: cp.ndarray	cp.ndarray	Extract Y channel as grayscale
ycbcr_full_to_legal(image)	image: cp.ndarray	cp.ndarray	Full to Legal range conversion
ycbcr_legal_to_full(image)	image: cp.ndarray	cp.ndarray	Legal to Full range conversion
yuv420p_to_ycbcr444(yuv420_data, width, height, interpolation)	yuv420_data: cp.ndarray, width: int, height: int, interpolation: int = 1	cp.ndarray	YUV 4:2:0 to YCbCr 4:4:4
yuv420p_to_ycbcr444_cp(yuv420_data, width, height, interpolation)	yuv420_data: cp.ndarray, width: int, height: int, interpolation: int = 1	cp.ndarray	YUV 4:2:0 to YCbCr 4:4:4 (CuPy native)
yuv422p10le_to_ycbcr444(ycbcr422_data, width, height)	ycbcr422_data: cp.ndarray, width: int, height: int	cp.ndarray	10-bit YUV 4:2:2 to YCbCr 4:4:4
yuv422p10le_to_ycbcr444_cp(ycbcr422_data, width, height)	ycbcr422_data: cp.ndarray, width: int, height: int	cp.ndarray	10-bit YUV 4:2:2 to YCbCr 4:4:4 (CuPy)
uyvy422_to_ycbcr444(uyvy_data, height, width)	uyvy_data: cp.ndarray, height: int, width: int	cp.ndarray	UYVY 4:2:2 to YCbCr 4:4:4
uyvy422_to_ycbcr444_cp(uyvy_data, height, width)	uyvy_data: cp.ndarray, height: int, width: int	cp.ndarray	UYVY 4:2:2 to YCbCr 4:4:4 (CuPy native)
ndi_uyvy422_to_ycbcr444(uyvy_data)	uyvy_data: cp.ndarray	cp.ndarray	NDI UYVY to YCbCr 4:4:4
ndi_uyvy422_to_ycbcr444_cp(uyvy_data)	uyvy_data: cp.ndarray	cp.ndarray	NDI UYVY to YCbCr 4:4:4 (CuPy native)

ACES Color Pipeline

Function	Parameters	Returns	Description
rec709_to_aces2065_1(image, tonemap)	image: cp.ndarray | np.ndarray, tonemap: bool = True	cp.ndarray | np.ndarray	Rec.709 to ACES2065-1 IDT. Preserves input type (NumPy/CuPy). OCIO ACES 1.2 compliant
aces2065_1_to_rec709(image, tonemap)	image: cp.ndarray | np.ndarray, tonemap: bool = True	cp.ndarray | np.ndarray	ACES2065-1 to Rec.709 ODT. Preserves input type (NumPy/CuPy). OCIO ACES 1.2 compliant
aces2065_1_to_acescct(image)	image: cp.ndarray | np.ndarray	cp.ndarray | np.ndarray	ACES2065-1 (AP0) to ACEScct (AP1 log-encoded). Preserves input type
aces2065_1_to_acescg(image)	image: cp.ndarray | np.ndarray	cp.ndarray | np.ndarray	ACES2065-1 (AP0) to ACEScg (AP1 linear). Preserves input type
acescct_to_aces2065_1(image)	image: cp.ndarray | np.ndarray	cp.ndarray | np.ndarray	ACEScct (AP1 log-encoded) to ACES2065-1 (AP0). Preserves input type
acescg_to_aces2065_1(image)	image: cp.ndarray | np.ndarray	cp.ndarray | np.ndarray	ACEScg (AP1 linear) to ACES2065-1 (AP0). Preserves input type

3D LUT Processing

Function	Parameters	Returns	Description
read_lut(file_path, use_cache, cache_dir)	file_path: str, use_cache: bool = True, cache_dir: str = "cache"	cp.ndarray	Read .cube format LUT files
apply_lut(image, lut, interpolation)	image: cp.ndarray, lut: cp.ndarray, interpolation: int = 0	cp.ndarray	Apply 3D LUT (0=trilinear, 1=tetrahedral)
apply_lut_cp(image, lut, interpolation)	image: cp.ndarray, lut: cp.ndarray, interpolation: int = 0	cp.ndarray	Apply 3D LUT (CuPy native implementation)

🖼️ Draw Module (pixtreme.draw)

Function	Parameters	Returns	Description
circle(image, center_x, center_y, radius, color)	image: cp.ndarray, center_x: int, center_y: int, radius: int, color: tuple = (1.0, 1.0, 1.0)	cp.ndarray	Draw filled circle
rectangle(image, top_left_x, top_left_y, bottom_right_x, bottom_right_y, color)	image: cp.ndarray, top_left_x: int, top_left_y: int, bottom_right_x: int, bottom_right_y: int, color: tuple = (1.0, 1.0, 1.0)	cp.ndarray	Draw filled rectangle
put_text(image, text, org, font_face, font_scale, color, thickness, line_type, density)	image: cp.ndarray, text: str, org: tuple[int, int], font_face: int = cv2.FONT_HERSHEY_SIMPLEX, font_scale: float = 1.0, color: tuple = (1.0, 1.0, 1.0), thickness: int = 2, line_type: int = cv2.LINE_AA, density: float = 1.0	cp.ndarray	Draw text with supersampling
add_label(image, text, org, font_face, font_scale, color, thickness, line_type, label_size, label_color, label_align, density)	image: cp.ndarray, text: str, org: tuple = (0, 0), font_face: int = cv2.FONT_HERSHEY_SIMPLEX, font_scale: float = 1.0, color: tuple = (1.0, 1.0, 1.0), thickness: int = 2, line_type: int = cv2.LINE_AA, label_size: int = 20, label_color: tuple = (0.0, 0.0, 0.0), label_align: str = "bottom", density: float = 1.0	cp.ndarray	Add labeled banner
create_rounded_mask(dsize, mask_offsets, radius_ratio, density, blur_size, sigma)	dsize: tuple = (512, 512), mask_offsets: tuple = (0.1, 0.1, 0.1, 0.1), radius_ratio: float = 0.1, density: int = 1, blur_size: int = 0, sigma: float = 1.0	cp.ndarray	Create rounded rectangle mask

🔨 Filter Module (pixtreme.filter)

Function	Parameters	Returns	Description
gaussian_blur(image, kernel_size, sigma, kernel)	image: cp.ndarray, kernel_size: int, sigma: float, kernel: cp.ndarray or None = None	cp.ndarray	Apply Gaussian blur
get_gaussian_kernel(ksize, sigma)	ksize: int, sigma: float	cp.ndarray	Generate 1D Gaussian kernel
GaussianBlur(kernel_size, sigma)	Class - kernel_size: int, sigma: float	Class instance	Gaussian blur filter class

🔄 Transform Module (pixtreme.transform)

Image Operations

Function	Parameters	Returns	Description
resize(src, dsize, fx, fy, interpolation)	src: cp.ndarray, dsize: tuple[int, int] or None = None, fx: float or None = None, fy: float or None = None, interpolation: int = INTER_AUTO	cp.ndarray	Resize image with 11 interpolation methods
affine_transform(src, M, dsize, flags)	src: cp.ndarray, M: cp.ndarray, dsize: tuple, flags: int = INTER_AUTO	cp.ndarray	Apply affine transformation matrix
get_inverse_matrix(M)	M: cp.ndarray	cp.ndarray	Calculate inverse transformation matrix
crop_from_kps(image, kps, size)	image: cp.ndarray, kps: cp.ndarray, size: int = 512	tuple[cp.ndarray, cp.ndarray]	Crop image based on keypoints
erode(image, kernel_size, kernel, border_value)	image: cp.ndarray, kernel_size: int, kernel: cp.ndarray or None = None, border_value: float = 0.0	cp.ndarray	Morphological erosion
create_erode_kernel(kernel_size)	kernel_size: int	cp.ndarray	Create erosion kernel
stack_images(images, axis)	images: list[cp.ndarray], axis: int = 0	cp.ndarray	Stack multiple images
subsample_image(image, factor)	image: cp.ndarray, factor: int	cp.ndarray	Fast downsampling
subsample_image_back(image, original_shape, factor)	image: cp.ndarray, original_shape: tuple, factor: int	cp.ndarray	Upsample back to original
tile_image(input_image, tile_size, overlap)	input_image: cp.ndarray, tile_size: int = 128, overlap: int = 16	tuple[list[cp.ndarray], tuple]	Split image into tiles
merge_tiles(tiles, original_shape, padded_shape, scale, tile_size, overlap)	tiles: list[cp.ndarray], original_shape: tuple[int, int, int], padded_shape: tuple[int, int, int], scale: int, tile_size: int = 128, overlap: int = 16	cp.ndarray	Merge tiles with blending
add_padding(input_image, patch_size, overlap)	input_image: cp.ndarray, patch_size: int = 128, overlap: int = 16	cp.ndarray	Add padding for tiling
create_gaussian_weights(size, sigma)	size: int, sigma: int	cp.ndarray	Create Gaussian weight map

Interpolation Constants

• INTER_NEAREST = 0 - Nearest neighbor
• INTER_LINEAR = 1 - Bilinear
• INTER_CUBIC = 2 - Bicubic
• INTER_AREA = 3 - Area-based resampling
• INTER_LANCZOS4 = 4 - Lanczos (8x8)
• INTER_AUTO = 5 - Auto-select
• INTER_MITCHELL = 6 - Mitchell-Netravali
• INTER_B_SPLINE = 7 - B-spline
• INTER_CATMULL_ROM = 8 - Catmull-Rom
• INTER_LANCZOS2 = 9 - Lanczos (4x4)
• INTER_LANCZOS3 = 10 - Lanczos (6x6)

📁 I/O Module (pixtreme.io)

Function	Parameters	Returns	Description
imread(input_path, dtype)	input_path: str, dtype: str = "fp32"	cp.ndarray	Read image to GPU (defaults to float32, always returns BGR)
imwrite(output_path, image, params)	output_path: str, image: cp.ndarray or np.ndarray, params: list or None = None	bool	Write image with OpenCV-compatible params (JPEG quality, PNG compression, etc.)
imencode(image, ext, params)	image: cp.ndarray, ext: str = ".png", params: list or None = None	bytes	Encode image to memory with OpenCV-compatible params
imdecode(src, dtype)	src: bytes, dtype: str = "fp32"	cp.ndarray	Decode image from memory (always returns BGR)
imshow(title, image, scale, is_rgb)	title: str, image: np.ndarray or cp.ndarray, scale: float = 1.0, is_rgb: bool = False	None	Display image
waitkey(delay)	delay: int	int	Wait for keyboard input
destroy_all_windows()	-	None	Close all OpenCV windows

🤖 Upscale Module (pixtreme.upscale)

Class	Constructor Parameters	Method	Description
OnnxUpscaler	model_path: str or None = None, model_bytes: bytes or None = None, device_id: int = 0, provider_options: list or None = None	get(image: cp.ndarray) -> cp.ndarray	ONNX Runtime upscaling
TrtUpscaler	model_path: str or None = None, model_bytes: bytes or None = None, device_id: int = 0	get(image: cp.ndarray) -> cp.ndarray	TensorRT optimized upscaling
TorchUpscaler	model_path: str or None = None, model_bytes: bytes or None = None, device: str = "cuda"	get(image: cp.ndarray) -> cp.ndarray	PyTorch native upscaling

🔧 Utils Module (pixtreme.utils)

Type Conversions

Function	Parameters	Returns	Description
to_uint8(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert to 8-bit (0-255)
to_uint16(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert to 16-bit unsigned
to_float16(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert to 16-bit float
to_float32(image, clip)	image: np.ndarray or cp.ndarray, clip: bool = True	np.ndarray or cp.ndarray	Convert to 32-bit float. If clip=True, clamps to [0,1]. If clip=False, preserves values outside [0,1] for scene-referred workflows (ACES)
to_float64(image)	image: np.ndarray or cp.ndarray	np.ndarray or cp.ndarray	Convert to 64-bit float
to_dtype(image, dtype)	image: np.ndarray or cp.ndarray, dtype: str	np.ndarray or cp.ndarray	Convert to specified dtype

Framework Interoperability

Function	Parameters	Returns	Description
to_cupy(image)	image: np.ndarray or torch.Tensor or nvimgcodec.Image	cp.ndarray	Convert to CuPy array
to_numpy(image)	image: cp.ndarray or torch.Tensor or nvimgcodec.Image	np.ndarray	Convert to NumPy array
to_tensor(image, device)	image: np.ndarray or cp.ndarray, device: str or torch.device or None = None	torch.Tensor	Convert to PyTorch tensor

Batch Processing

Function	Parameters	Returns	Description
guess_image_layout(image)	image: np.ndarray or cp.ndarray	str	Detect image layout (HW, HWC, CHW, etc.)
image_to_batch(image, size, scalefactor, mean, swap_rb, layout)	image: cp.ndarray, size: int or tuple[int, int] or None = None, scalefactor: float or None = None, mean: float or tuple or None = None, swap_rb: bool = True, layout: str = "HWC"	cp.ndarray	Convert single image to batch
images_to_batch(images, size, scalefactor, mean, swap_rb, layout)	images: list[cp.ndarray], size: int or tuple[int, int] or None = None, scalefactor: float or None = None, mean: float or tuple or None = None, swap_rb: bool = True, layout: str = "HWC"	cp.ndarray	Convert images to batch format
batch_to_images(batch, scalefactor, mean, swap_rb, layout)	batch: cp.ndarray, scalefactor: float or tuple or None = None, mean: float or tuple or None = None, swap_rb: bool = True, layout: str = "NCHW"	list[cp.ndarray]	Convert batch to images

Model Conversion

Function	Parameters	Returns	Description
check_torch_model(model_path)	model_path: str	bool	Validate PyTorch model
check_onnx_model(model_path)	model_path: str	bool	Validate ONNX model
torch_to_onnx(model_path, onnx_path, input_shape, opset_version, precision, dynamic_axes, device)	model_path: str, onnx_path: str, input_shape: tuple = (1,3,1080,1920), opset_version: int = 20, precision: str = "fp32", dynamic_axes: dict or None = None, device: str = "cuda"	None	Convert PyTorch to ONNX
onnx_to_onnx_dynamic(input_path, output_path, opset, irver)	input_path: str, output_path: str, opset: int or None = None, irver: int or None = None	None	Add dynamic shape support
onnx_to_trt(onnx_path, engine_path, precision, workspace)	onnx_path: str, engine_path: str, precision: str = "fp16", workspace: int = 1<<30	None	Convert ONNX to TensorRT
onnx_to_trt_dynamic_shape(onnx_path, engine_path, precision, workspace)	onnx_path: str, engine_path: str, precision: str = "fp16", workspace: int = 1<<30	None	TensorRT with dynamic shapes
onnx_to_trt_fixed_shape(onnx_path, engine_path, precision, workspace, input_shape)	onnx_path: str, engine_path: str, precision: str = "fp16", workspace: int = 1<<30, input_shape: tuple = (1,3,1080,1920)	None	TensorRT with fixed shape

Performance Notes

• All color conversion operations use optimized CUDA kernels
• Supports both legal range (16-235) and full range (0-255) for video processing
• 10-bit precision support for professional video workflows
• Zero-copy tensor sharing via DLPack for framework interoperability
• Batch processing support for multiple images

License

pixtreme is distributed under the MIT License (see LICENSE).

Authors

minamik (@minamikik)

Acknowledgments

sync.dev