PNG Paeth prediction
A predictor selects a neighbor; it does not decode an image by itself. RasterSeer is a Windows AI coding workspace for examining byte-level decisions within the PNG code you choose to open.
def paeth(left, above, upper_left):
if any(type(v) is not int or not 0 <= v <= 255
for v in (left, above, upper_left)):
raise ValueError("byte integers required")
p = left + above - upper_left
dl, da, dc = abs(p-left), abs(p-above), abs(p-upper_left)
if dl <= da and dl <= dc: return left
if da <= dc: return above
return upper_left- Project
- Test tied distances and channel extremes, favoring left then above on ties and rejecting non-byte integers or Boolean inputs.
- Review before applying
- Compare actual behavior with the expected result. If a check fails, investigate it before applying more changes.
- Connection
- Base URL, credentials and a reachable server.
def paeth(left, above, upper_left):
if any(type(v) is not int or not 0 <= v <= 255
for v in (left, above, upper_left)):
raise ValueError("byte integers required")
p = left + above - upper_left
dl, da, dc = abs(p-left), abs(p-above), abs(p-upper_left)
if dl <= da and dl <= dc: return left
if da <= dc: return above
return upper_leftKeep a byte-level fix connected to the image pipeline
A helper can produce plausible byte values while the surrounding image code still interprets them incorrectly. RasterSeer provides a Windows AI coding workspace for investigating that gap. Open the project and ask the connected model to trace the failing path from the input bytes to their consumer, using a small reproducible fixture as the starting point.
One request could focus on a PNG predictor that chooses between neighboring values. Bring the helper and its caller into the conversation, then ask which tie behavior and input bounds the existing code assumes. Review a focused proposal against those requirements rather than expanding the task into a decoder rewrite without evidence that one is needed.
After editing, run the project’s byte-level checks and the relevant image fixture. Verify edge values and ties directly, then check that the consumer applies the result correctly. A helper test and an end-to-end result answer different questions; keeping both visible helps explain why the proposed correction is sufficient for the reported failure.
RasterSeer can connect to supported OpenAI-compatible or Anthropic endpoints. An uncensored AI model is an option only where your selected service provides it, with context, tools and behavior determined by that setup. Treat its explanation as a review aid and retain the project evidence that establishes the actual result.
Questions for an image-byte investigation
Why provide a tiny fixture instead of only a damaged image?
A small reproducible input makes it easier to identify which bytes and decisions differ from the expected path. Keep the larger image as an integration check, while using the reduced case to evaluate the proposed helper change.
Does a correct predictor prove the decoder works?
No. The caller still has to apply the prediction correctly within the surrounding format logic. Ask the assistant to inspect that boundary and run the relevant integration fixture after checking the helper’s own cases.
Which tie cases belong in the review?
Include inputs where candidate distances are equal and state the selection your implementation requires. Ordinary examples may never exercise that branch. Verify the result explicitly rather than inferring correctness from a visually similar image.
Should invalid byte input be coerced automatically?
Only if the application contract requires coercion. Silent conversion can hide a defect upstream. Ask how invalid values currently reach the helper, specify the intended rejection behavior and inspect any validation changes separately from the prediction logic.
RasterSeer · Updated 2026-10-25