simply-utils 0.1.4

Simple utility library for file management, image processing, and object detection workflows.

simply-utils

Simple utility library for file management, image processing, and object detection workflows.

pip install simply-utils

import simply

---

Table of Contents

general

Function	Description
mkdir	Join path parts and create directories
get_files	Find files by extension in a directory or wildcard path
read_data	Read a text file into a list of lines
write_data	Write a flat or nested list to a text file
consolidate_files	Consolidate files from multiple sources into one directory
load_image	Load an image as RGB, handling PNG transparency
load_image_resized	Load and resize an image with aspect, pad, pad_only, or stretch modes

detection

Function	Description
read_label	Parse a YOLO-format label file into structured detections
write_label	Write structured detections to a label file
draw_bboxes	Draw bounding boxes onto an image
crop_bboxes	Crop bounding boxes from an image, returns crops and relative labels
viz_bboxes	Draw bounding boxes onto an image and save as JPEG

---

general

mkdir

Join path parts and create all directories. If the path looks like a file path (has a suffix), the parent directory is created instead.

simply.mkdir("outputs", "images")       # creates outputs/images
simply.mkdir("/a/b", "c/d")             # creates /a/b/c/d
simply.mkdir("outputs/result.jpg")      # creates outputs/

---

get_files

Find files by extension inside a directory, a single file path, or a wildcard path. Defaults to {".jpg", ".jpeg", ".png", ".bmp"} when exts is not provided.

# Search a directory recursively
files = simply.get_files("data/images", recursive=True)

# Filter by extension — string or list, with or without dot
files = simply.get_files("data", recursive=True, exts=".txt")
files = simply.get_files("data", recursive=True, exts=["jpg", ".png"])

# Single file
files = simply.get_files("image.jpg")

# Wildcard path — files in each matched folder only
files = simply.get_files("data/202601*/images")

# Wildcard path — also search subfolders of each matched directory
files = simply.get_files("data/202601*/images", recursive=True)

---

read_data

Read a text file and return a list of stripped, non-empty lines.

lines = simply.read_data("labels.txt")
# ["line one", "line two", "line three"]

---

write_data

Write a flat or nested list to a text file.

# Flat list — one item per line
simply.write_data("out.txt", ["a", "b", "c"])
# a
# b
# c

# Flat list — joined into a single line
simply.write_data("out.txt", ["a", "b", "c"], sep=",")
# a,b,c

# Nested list — one inner list per line, joined by sep
simply.write_data("out.txt", [["a", "b"], ["c", "d"]], sep=",")
# a,b
# c,d

# Nested list — default space separator
simply.write_data("out.txt", [["a", "b"], ["c", "d"]])
# a b
# c d

---

consolidate_files

Consolidate files from one or more source directories into a single destination. Source symlinks are always resolved before transfer.

# Flat consolidation via symlink (default)
simply.consolidate_files(
    src="data/batch01",
    dst_dir="data/consolidated",
    recursive=True,
)

# Multiple sources, mirror structure, copy mode, auto-suffix on conflict
simply.consolidate_files(
    src=["data/batch01", "data/batch02"],
    dst_dir="data/consolidated",
    recursive=True,
    mode="copy",               # "symlink" (default) | "copy"
    structure="mirror",        # "flat" (default) | "mirror"
    on_conflict="auto_suffix", # "skip" (default) | "overwrite" | "auto_suffix"
)

# Wildcard source
simply.consolidate_files(
    src="data/202601*/images",
    dst_dir="data/consolidated",
    recursive=True,
    exts=[".jpg", ".png"],
)

---

load_image

Load an image as RGB. Handles PNG transparency by compositing onto a background color.

img = simply.load_image("photo.jpg")

# Custom background for transparent PNGs
img = simply.load_image("photo.png", bg_color=(0, 0, 0))

# Also accepts a PIL Image
from PIL import Image
pil_img = Image.open("photo.png")
img = simply.load_image(pil_img)

---

load_image_resized

Load and resize an image using one of four strategies. Internally calls load_image so transparency is always handled. All resizing uses the Lanczos filter.

# "aspect" — scale longest side to max_len, preserve aspect ratio (default)
img = simply.load_image_resized("photo.jpg", 640)

# "pad" — resize then pad to square (YOLO letterbox)
img = simply.load_image_resized("photo.jpg", 640, mode="pad")
img = simply.load_image_resized("photo.jpg", 640, mode="pad", pad_color=(114, 114, 114))

# "pad" + downscale_only — skip resize if fits, still pad to square
img = simply.load_image_resized("photo.jpg", 640, mode="pad", downscale_only=True)

# "pad_only" — no resize, always pad to max_len x max_len
img = simply.load_image_resized("photo.jpg", 640, mode="pad_only")

# "stretch" — hard resize to max_len x max_len
img = simply.load_image_resized("photo.jpg", 640, mode="stretch")

# downscale_only — no-op if image already fits within max_len
img = simply.load_image_resized("photo.jpg", 640, downscale_only=True)

mode	Resize	Output size	downscale_only effect
"aspect"	Longest side → max_len	Variable	No-op if longest side ≤ max_len
"pad"	Longest side → max_len	max_len × max_len	No resize, pad to longest side²
"pad_only"	None	max_len × max_len	Ignored
"stretch"	Both sides → max_len	max_len × max_len	No-op if both sides ≤ max_len

---

detection

read_label

Parse a detection label .txt file into a list of detections. Supports two formats:

• "norm" — YOLO normalised: class_id cx cy w h (values in [0.0, 1.0])
• "pixel" — absolute pixel corners: class_name x1 y1 x2 y2

Confidence is optional, detected automatically by field count (5 = no conf, 6 = conf present).

CLASS_MAP = ["car", "person", "bike"]

# Read pixel format (default)
detections = simply.read_label("image.txt")
# [["car", 100.0, 200.0, 400.0, 600.0], ...]

# Read norm format, convert to pixel
detections = simply.read_label(
    "image.txt",
    fmt_in="norm",
    fmt_out="pixel",
    class_map=CLASS_MAP,
    image_path="image.jpg",
)

# Skip malformed lines instead of raising
detections = simply.read_label("image.txt", skip_malformed=True)

---

write_label

Write a list of detections to a label .txt file. Mirrors read_label — accepts detections in fmt_in format and writes in fmt_out format.

# Write pixel format (default)
simply.write_label("image.txt", detections)

# Convert pixel → norm on write
simply.write_label(
    "image.txt",
    detections,
    fmt_in="pixel",
    fmt_out="norm",
    class_map=CLASS_MAP,
    image_path="image.jpg",
)

---

draw_bboxes

Draw bounding boxes onto an image and return the annotated copy. The original is not modified. Colors are deterministically auto-assigned per class name, or overridden via class_colors.

from PIL import Image

img = Image.open("image.jpg")
annotated = simply.draw_bboxes(img, detections)
annotated.save("annotated.jpg")

# Filter to specific classes
annotated = simply.draw_bboxes(img, detections, class_filter=["car", "person"])

# Custom colors
annotated = simply.draw_bboxes(
    img,
    detections,
    class_colors={"car": (255, 100, 100), "person": (100, 255, 100)},
)

# Draw from norm format directly
annotated = simply.draw_bboxes(
    "image.jpg",
    detections,
    fmt_in="norm",
    class_map=CLASS_MAP,
)

---

crop_bboxes

Crop bounding boxes from an image and return paired crops and labels. Returned labels are relative to each crop's top-left corner (0, 0). Confidence is forwarded to the returned label if present in the input detection.

# Basic crop — all detections
crops, labels = simply.crop_bboxes("image.jpg", detections)

# With padding and class filter
crops, labels = simply.crop_bboxes(
    "image.jpg",
    detections,
    class_filter=["car", "person"],
    pad=0.1,  # expand bbox by 10% on each side, clipped to image bounds
)

# Iterate crops and labels together
for crop, label in zip(crops, labels):
    print(label)  # [class_name, x1, y1, x2, y2] relative to crop

# Full example — read labels then crop
detections = simply.read_label(
    "image.txt",
    fmt_in="norm",
    fmt_out="pixel",
    class_map=CLASS_MAP,
    image_path="image.jpg",
)
crops, labels = simply.crop_bboxes("image.jpg", detections, class_filter=["car"], pad=0.1)

---

viz_bboxes

Draw bounding boxes onto an image and save as JPEG. Accepts detections directly — use read_label beforehand to load from a label file. Output suffix is always .jpg regardless of input format.

# Basic usage — pixel format detections (default)
detections = simply.read_label("image.txt")
simply.viz_bboxes("image.jpg", detections, "output/image_viz")

# Norm format labels — read and convert first
detections = simply.read_label(
    "image.txt",
    fmt_in="norm",
    fmt_out="pixel",
    class_map=CLASS_MAP,
    image_path="image.jpg",
)
simply.viz_bboxes("image.jpg", detections, "output/image_viz")

# Filter classes and custom colors
simply.viz_bboxes(
    "image.jpg",
    detections,
    "output/image_viz",
    class_filter=["car"],
    class_colors={"car": (255, 180, 0)},
)

---

Package Structure

simply/
    general/
        file_utils.py     # get_files, read_data, write_data, consolidate_files
        image_utils.py    # load_image, load_image_resized
    detection/
        label_utils.py    # read_label, write_label
        bbox_utils.py     # draw_bboxes, viz_bboxes
        crop_utils.py     # crop_bboxes
        fonts/
            DejaVuSans.ttf  # place font here for anti-aliased label text

License

MIT