Photo2Pixel

---

English | 简体中文

Online Tool | Colab | Tutorial

photo2pixel is an algorithm converting photo into pixel art. There is an online converter coding.tools/photo2pixel . you can try different combination of pixel size and edge threshold to get the best result.

Prerequisites

• python3
• pytorch (for algorithm implementation)
• pillow (for image file io)
• onnx and onnxruntime (for single-file ONNX export/inference)

Tutorial

---

photo2pixel is implemented with Pytorch, the easiest way to run it is Colab, or you can run it with command as bellow:

# use default param
python convert.py --input ./images/example_input_mountain.jpg

# or use custom param
python convert.py --kernel_size 12 --pixel_size 12 --edge_thresh 128

Export one configurable ONNX file and reuse it for different settings:

python export_onnx.py --output ./photo2pixel.onnx
python convert_onnx.py --model ./photo2pixel.onnx --kernel_size 12 --pixel_size 12 --edge_thresh 128
python convert_onnx.py --model ./photo2pixel.onnx --kernel_size 25 --pixel_size 8 --edge_thresh 80

The exported ONNX file has runtime inputs for kernel_size, pixel_size, and edge_thresh, so one file can be reused across config combinations.

Parameter	Description	Range	Default
input	input image path	/	./images/example_input_mountain.jpg
output	output image path	/	./result.png
kernel_size	larger kernel size means smooth color transition	unlimited	10
pixel_size	individual pixel size	unlimited	16
edge_thresh	the black line in edge region, lower edge threshold means more black line	0~255	100

updated by openclaw at 20260312