visual-logging

A simple way to generate beautiful html logs with embedded images for CV purposes.

visual-logging piggybacks on the standard logging module and allows you to use the sick power of logging to debug your computer vision application on a whole new level: OpenCV images (well, technically numpy arrays), PIL images and matplotlib figures get embedded into your html logs as inline images.

You can read about it in detail in a great blog post visual-logging, my new favorite tool for debugging OpenCV and Python apps written by Adrian Rosebrock.

Installation

pip install visual-logging

No extra dependencies — whichever of OpenCV, PIL/Pillow and matplotlib you already have installed are picked up automatically (numpy arrays render through PIL when OpenCV isn't around). Requires Python 3.9+.

Usage example (see demo.py)

import logging
from vlogging import HTMLFileHandler, VisualRecord

import cv2  # or PIL.Image, or matplotlib — whatever you use

logger = logging.getLogger("demo")
logger.setLevel(logging.DEBUG)
logger.addHandler(HTMLFileHandler("test.html", title="My debug log"))

cv_image = cv2.imread("lenna.jpg")

logger.debug(VisualRecord(
    "Hello from OpenCV", cv_image, "This is an OpenCV image", fmt="png"))

# Mix images of different origins in one record, downscale the big ones
# to at most 320x240 to keep the log small:
logger.warning(VisualRecord(
    "Hello from all", [cv_image, pil_image, mpl_figure],
    fmt="png", max_size=(320, 240)))

# Ordinary log calls work too, and land in the same page:
logger.info("Processed frame %d", 42)

logging.shutdown()  # flushes and closes the html file

Open test.html in a browser and enjoy: HTMLFileHandler writes a styled, self-contained page — records are color-coded by log level with timestamps and logger names, plain messages are escaped, and exceptions logged with logger.exception(...) include their traceback. Records are flushed as they happen, so you can watch the page mid-run.

Everything composes the stdlib way: HTMLFileHandler is a logging.FileHandler that installs a VisualFormatter (a logging.Formatter) by default — use either piece on its own if you prefer. Passing a VisualRecord to a plain FileHandler still produces bare html fragments, exactly as in 1.x.

In Jupyter, a VisualRecord displays itself inline — no logging setup needed:

VisualRecord("Detected edges", edges_img, "Canny output")

VisualRecord arguments

Changelog

2.0

• Modern packaging (pyproject.toml), Python 3.9+ only, py.typed type hints
• New HTMLFileHandler + VisualFormatter: styled, self-contained html pages with level colors, timestamps, escaped plain-text records and tracebacks
• New max_size option to downscale embedded images (all renderers)
• VisualRecord displays inline in Jupyter notebooks
• numpy arrays render via PIL when OpenCV is not installed
• Embedded images use loading="lazy", so huge logs open fast
• matplotlib support no longer relies on the deprecated pylab module
• Tests run on GitHub Actions against Python 3.9–3.13; releases publish to PyPI from tags via trusted publishing