omnipart's picture
init
491eded
"""
General utility functions for the TRELLIS project.
This module provides a collection of utility functions organized into several categories:
- Dictionary utilities: Operations for merging, traversal, reduction, and flattening of dictionaries
- Context management: Tools for managing nested context managers
- Image utilities: Functions for creating image grids and annotating images
- Debug utilities: Tools for numerical comparison and tolerance checking
- Print utilities: Helpers for text formatting and display
"""
import re
import numpy as np
import cv2
import torch
import contextlib
# Dictionary utils
def _dict_merge(dicta, dictb, prefix=''):
"""
Merge two dictionaries recursively with conflict detection.
Args:
dicta: First dictionary to merge
dictb: Second dictionary to merge
prefix: Used for error reporting to track nested keys
Returns:
A new merged dictionary
Raises:
ValueError: If the same key exists in both dictionaries but has different types
"""
assert isinstance(dicta, dict), 'input must be a dictionary'
assert isinstance(dictb, dict), 'input must be a dictionary'
dict_ = {}
# Get all unique keys from both dictionaries
all_keys = set(dicta.keys()).union(set(dictb.keys()))
for key in all_keys:
if key in dicta.keys() and key in dictb.keys():
# If key exists in both, recursively merge if both are dictionaries
if isinstance(dicta[key], dict) and isinstance(dictb[key], dict):
dict_[key] = _dict_merge(dicta[key], dictb[key], prefix=f'{prefix}.{key}')
else:
# Raise error for conflicting non-dictionary values
raise ValueError(f'Duplicate key {prefix}.{key} found in both dictionaries. Types: {type(dicta[key])}, {type(dictb[key])}')
elif key in dicta.keys():
# Copy values from first dictionary
dict_[key] = dicta[key]
else:
# Copy values from second dictionary
dict_[key] = dictb[key]
return dict_
def dict_merge(dicta, dictb):
"""
Merge two dictionaries.
This is the public interface that wraps _dict_merge with default prefix.
"""
return _dict_merge(dicta, dictb, prefix='')
def dict_foreach(dic, func, special_func={}):
"""
Recursively apply a function to all non-dictionary leaf values in a dictionary.
Args:
dic: Dictionary to process
func: Default function to apply to each leaf value
special_func: Dictionary mapping keys to special functions for specific keys
Returns:
Transformed dictionary with function applied to all leaf values
"""
assert isinstance(dic, dict), 'input must be a dictionary'
for key in dic.keys():
if isinstance(dic[key], dict):
# Recursively process nested dictionaries
dic[key] = dict_foreach(dic[key], func)
else:
# Apply special function if key is in special_func, otherwise use default
if key in special_func.keys():
dic[key] = special_func[key](dic[key])
else:
dic[key] = func(dic[key])
return dic
def dict_reduce(dicts, func, special_func={}):
"""
Reduce a list of dictionaries. Leaf values must be scalars.
Args:
dicts: List of dictionaries to reduce
func: Default reduction function (takes a list of values, returns single value)
special_func: Dictionary mapping keys to special reduction functions
Returns:
A single merged dictionary with values reduced according to the provided functions
"""
assert isinstance(dicts, list), 'input must be a list of dictionaries'
assert all([isinstance(d, dict) for d in dicts]), 'input must be a list of dictionaries'
assert len(dicts) > 0, 'input must be a non-empty list of dictionaries'
# Collect all unique keys across all dictionaries
all_keys = set([key for dict_ in dicts for key in dict_.keys()])
reduced_dict = {}
for key in all_keys:
# Extract values for this key from all dictionaries
vlist = [dict_[key] for dict_ in dicts if key in dict_.keys()]
if isinstance(vlist[0], dict):
# Recursively reduce nested dictionaries
reduced_dict[key] = dict_reduce(vlist, func, special_func)
else:
# Apply special function if key is in special_func, otherwise use default
if key in special_func.keys():
reduced_dict[key] = special_func[key](vlist)
else:
reduced_dict[key] = func(vlist)
return reduced_dict
def dict_any(dic, func):
"""
Check if any value in the dictionary satisfies the given predicate function.
Args:
dic: Dictionary to check
func: Predicate function that returns True/False for each leaf value
Returns:
True if any leaf value satisfies the predicate, False otherwise
dict any time: {'step': 16.795613527297974, 'elapsed': 16.795613527297974}
dict any step: 16.795613527297974
dict any elapsed: 16.795613527297974
dict any loss: {'bin_3': {'mse': nan}, 'bin_5': {'mse': nan}, 'mse': nan, 'loss': nan, 'bin_4': {'mse': nan}, 'bin_7': {'mse': nan}, 'bin_8': {'mse': nan}}
dict any bin_3: {'mse': nan}
dict any mse: nan
"""
assert isinstance(dic, dict), 'input must be a dictionary'
for key in dic.keys():
# print(f"dict any {key}: {dic[key]}")
if isinstance(dic[key], dict):
# Recursively check nested dictionaries
if dict_any(dic[key], func):
return True
else:
# Check current value against predicate
if func(dic[key]):
return True
return False
def dict_all(dic, func):
"""
Check if all values in the dictionary satisfy the given predicate function.
Args:
dic: Dictionary to check
func: Predicate function that returns True/False for each leaf value
Returns:
True if all leaf values satisfy the predicate, False otherwise
"""
assert isinstance(dic, dict), 'input must be a dictionary'
for key in dic.keys():
if isinstance(dic[key], dict):
# Recursively check nested dictionaries
if not dict_all(dic[key], func):
return False
else:
# Check current value against predicate
if not func(dic[key]):
return False
return True
def dict_flatten(dic, sep='.'):
"""
Flatten a nested dictionary into a dictionary with no nested dictionaries.
Args:
dic: Dictionary to flatten
sep: Separator string used to join key levels in the flattened dictionary
Returns:
A flattened dictionary with compound keys joined by the separator
"""
assert isinstance(dic, dict), 'input must be a dictionary'
flat_dict = {}
for key in dic.keys():
if isinstance(dic[key], dict):
# Recursively flatten nested dictionaries and prefix with current key
sub_dict = dict_flatten(dic[key], sep=sep)
for sub_key in sub_dict.keys():
flat_dict[str(key) + sep + str(sub_key)] = sub_dict[sub_key]
else:
# Copy leaf values directly
flat_dict[key] = dic[key]
return flat_dict
# Context utils
@contextlib.contextmanager
def nested_contexts(*contexts):
"""
Create a single context manager from multiple context manager factories.
This utility allows combining multiple context managers into a single
context manager, simplifying code that needs to use multiple contexts.
Args:
*contexts: Context manager factory functions (functions that return context managers)
Yields:
Control to the inner block when all contexts are entered
"""
with contextlib.ExitStack() as stack:
for ctx in contexts:
stack.enter_context(ctx())
yield
# Image utils
def make_grid(images, nrow=None, ncol=None, aspect_ratio=None):
"""
Arrange multiple images into a grid.
Args:
images: List of images to arrange
nrow: Number of rows (calculated if not provided)
ncol: Number of columns (calculated if not provided)
aspect_ratio: Desired width/height ratio for the grid layout
Returns:
A single image containing the grid of input images
"""
num_images = len(images)
# Calculate grid dimensions if not explicitly provided
if nrow is None and ncol is None:
if aspect_ratio is not None:
# Calculate rows to achieve desired aspect ratio
nrow = int(np.round(np.sqrt(num_images / aspect_ratio)))
else:
# Default to a roughly square grid
nrow = int(np.sqrt(num_images))
ncol = (num_images + nrow - 1) // nrow
elif nrow is None and ncol is not None:
# Calculate rows based on fixed columns
nrow = (num_images + ncol - 1) // ncol
elif nrow is not None and ncol is None:
# Calculate columns based on fixed rows
ncol = (num_images + nrow - 1) // nrow
else:
assert nrow * ncol >= num_images, 'nrow * ncol must be greater than or equal to the number of images'
# Create empty grid with appropriate dimensions
if images[0].ndim == 2:
# Grayscale images
grid = np.zeros((nrow * images[0].shape[0], ncol * images[0].shape[1]), dtype=images[0].dtype)
else:
# Color images
grid = np.zeros((nrow * images[0].shape[0], ncol * images[0].shape[1], images[0].shape[2]), dtype=images[0].dtype)
# Place each image in the grid
for i, img in enumerate(images):
row = i // ncol
col = i % ncol
grid[row * img.shape[0]:(row + 1) * img.shape[0], col * img.shape[1]:(col + 1) * img.shape[1]] = img
return grid
def notes_on_image(img, notes=None):
"""
Add text notes to an image by padding the bottom and adding text.
Args:
img: Input image
notes: Text to add at the bottom of the image
Returns:
Image with notes added
"""
# Add padding at the bottom for the notes
img = np.pad(img, ((0, 32), (0, 0), (0, 0)), 'constant', constant_values=0)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
if notes is not None:
# Add text to the padded area
img = cv2.putText(img, notes, (0, img.shape[0] - 4), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 1)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img
def save_image_with_notes(img, path, notes=None):
"""
Save an image with optional text notes at the bottom.
Args:
img: Input image (numpy array or PyTorch tensor)
path: File path to save the image
notes: Optional text to add at the bottom of the image
"""
# Convert PyTorch tensor to numpy if needed
if isinstance(img, torch.Tensor):
img = img.cpu().numpy().transpose(1, 2, 0)
# Scale floating point images to 0-255 range
if img.dtype == np.float32 or img.dtype == np.float64:
img = np.clip(img * 255, 0, 255).astype(np.uint8)
# Add notes to the image
img = notes_on_image(img, notes)
# Save with proper color conversion for OpenCV
cv2.imwrite(path, cv2.cvtColor(img, cv2.COLOR_RGB2BGR))
# debug utils
def atol(x, y):
"""
Absolute tolerance - computes absolute difference between x and y.
Useful for numerical comparisons when absolute error matters.
Args:
x, y: Tensors to compare
Returns:
Absolute difference |x - y|
"""
return torch.abs(x - y)
def rtol(x, y):
"""
Relative tolerance - computes relative difference between x and y.
Useful for numerical comparisons when relative error matters,
especially when comparing values of different magnitudes.
Args:
x, y: Tensors to compare
Returns:
Relative difference |x - y| / max(|x|, |y|)
"""
return torch.abs(x - y) / torch.clamp_min(torch.maximum(torch.abs(x), torch.abs(y)), 1e-12)
# print utils
def indent(s, n=4):
"""
Indent a multi-line string.
Args:
s: Input string to indent
n: Number of spaces to add before each line (except the first)
Returns:
Indented string with all lines except the first indented by n spaces
"""
lines = s.split('\n')
for i in range(1, len(lines)):
lines[i] = ' ' * n + lines[i]
return '\n'.join(lines)