BOX filter missed some edge-case pixels

[AlttiRi]

What did you do?

I down-sample an image with BOX filter.

Here is the image:

table-51.png (width="2755" height="1837")

---

What did you expect to happen?

All pixels are counted while down-sampling.

The result should look like this (688x459, 459x306):

box-js-v3-table-51.png-688x459-49c214b-pil-like.png

---

box-js-v3-table-51.png-459x306-20d1f1f-pil-like.png

---

What actually happened?

Some edge-case pixels were skipped.

---

Downscaled 4 times to 688x459:

box-pil-table-51.png-688x459-49c7ade.png

missed a central vertical line.

---

Downscaled 6 times to 459x306:

box-pil-table-51.png-459x306-20d5e62.png

missed a central horizontal line.

---

What are your OS, Python and Pillow versions?

• OS: Win 10
• Python: 3.12
• Pillow: 12.3.0

file_path = "./table-51.png"
scale = 4  # scale(-down)

image = Image.open(file_path)
image_gray = image.convert("L")
wh = (int(image.width / scale), int(image.height / scale))
image = image_gray.resize(wh, Image.Resampling.BOX)

---

Search words: off-by-one error, range rounding.

[AlttiRi]

Seems, the problem exists not only with the central lines.

For example, here is the comparison of PIL's result and my BOX down-sampling implementation's (it's also with weight precomputing and with the fixed point math) result:

The visual diff of resampling (grayscaled) DR94LKg.jpeg (1923x2533) to 480x633 (down-scale is 4 times).