ewilliams28/openclaw
1
0
Fork 0
mirror of https://github.com/openclaw/openclaw.git synced 2026-08-12 21:53:00 -06:00
Code Issues Packages Projects Releases Wiki Activity

fix(document-extract): render PDF image fallback per page so multi-page scans don't starve later pages (#96390)

Browse Source 
* fix(document-extract): render PDF image fallback per page so multi-page scans don't starve later pages

clawpdf's mode:"images" extract applies a single maxPixels budget across
every page, so the first page consumes it and later pages collapse to ~1x1
PNGs that vision OCR models reject. Render each selected page in its own
extract() call so the pixel budget resets per page and every page yields a
usable image.

* fix(document-extract): preserve aggregate PDF render budget

---------

Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
This commit is contained in:
Wynne668
2026-06-25 16:37:47 +08:00
committed by GitHub
parent 42d0a1267e
commit d7dff3cbf4
2 changed files with 81 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
extensions/document-extract/document-extractor.test.ts
+50 -24
View File
@@ -55,20 +55,35 @@ describe("PDF document extractor", () => {
});
});
it("extracts text first and renders fallback images through clawpdf", async () => {
pdfDocument.extract.mockResolvedValueOnce({ text: "", images: [] }).mockResolvedValueOnce({
text: "",
images: [
{
type: "image",
bytes: Uint8Array.from(Buffer.from("png")),
mimeType: "image/png",
page: 1,
width: 10,
height: 10,
},
],
});
it("extracts text first and renders each fallback page with its own pixel budget", async () => {
pdfDocument.extract
.mockResolvedValueOnce({ text: "", images: [] })
.mockResolvedValueOnce({
text: "",
images: [
{
type: "image",
bytes: Uint8Array.from(Buffer.from("png1")),
mimeType: "image/png",
page: 1,
width: 5,
height: 10,
},
],
})
.mockResolvedValueOnce({
text: "",
images: [
{
type: "image",
bytes: Uint8Array.from(Buffer.from("png2")),
mimeType: "image/png",
page: 2,
width: 5,
height: 10,
},
],
});
const extractor = createPdfDocumentExtractor();
const result = await extractor.extract(request());
@@ -82,18 +97,24 @@ describe("PDF document extractor", () => {
maxPages: 2,
maxTextChars: 200_000,
});
// Each page renders in its own extract() call, with the aggregate pixel cap
// allocated across selected pages so later pages are not starved.
expect(pdfDocument.extract).toHaveBeenNthCalledWith(2, {
mode: "images",
maxPages: 2,
image: {
maxDimension: 10_000,
maxPixels: 100,
forms: true,
},
pages: [1],
image: { maxDimension: 10_000, maxPixels: 50, forms: true },
});
expect(pdfDocument.extract).toHaveBeenNthCalledWith(3, {
mode: "images",
pages: [2],
image: { maxDimension: 10_000, maxPixels: 50, forms: true },
});
expect(result).toEqual({
text: "",
images: [{ type: "image", data: "cG5n", mimeType: "image/png" }],
images: [
{ type: "image", data: "cG5nMQ==", mimeType: "image/png" },
{ type: "image", data: "cG5nMg==", mimeType: "image/png" },
],
});
expect(pdfDocument.destroy).toHaveBeenCalledTimes(1);
});
@@ -131,8 +152,9 @@ describe("PDF document extractor", () => {
expect(pdfDocument.destroy).not.toHaveBeenCalled();
});
it("filters selected pages before passing them to clawpdf", async () => {
it("filters selected pages and renders them one page per image call", async () => {
pdfDocument.extract
.mockResolvedValueOnce({ text: "", images: [] })
.mockResolvedValueOnce({ text: "", images: [] })
.mockResolvedValueOnce({ text: "", images: [] });
const extractor = createPdfDocumentExtractor();
@@ -141,11 +163,15 @@ describe("PDF document extractor", () => {
expect(pdfDocument.extract).toHaveBeenNthCalledWith(
1,
expect.objectContaining({ pages: [2, 1] }),
expect.objectContaining({ mode: "text", pages: [2, 1] }),
);
expect(pdfDocument.extract).toHaveBeenNthCalledWith(
2,
expect.objectContaining({ pages: [2, 1] }),
expect.objectContaining({ mode: "images", pages: [2] }),
);
expect(pdfDocument.extract).toHaveBeenNthCalledWith(
3,
expect.objectContaining({ mode: "images", pages: [1] }),
);
});
extensions/document-extract/document-extractor.ts
+31 -10
View File
@@ -83,17 +83,38 @@ async function extractPdfContent(
return { text, images: [] };
}
// clawpdf's image render budget (maxPixels) is shared across every page in one
// extract() call: the first page consumes it and later pages collapse to 1x1
// PNGs that vision models reject. Render each page separately, allocating the
// remaining aggregate budget across pages that still need rendering.
const imagePages =
pages ?? Array.from({ length: Math.min(pdf.pageCount, request.maxPages) }, (_, i) => i + 1);
try {
const imageResult = await pdf.extract({
mode: "images",
...pageSelection,
image: {
maxDimension: MAX_RENDER_DIMENSION,
maxPixels: request.maxPixels,
forms: true,
},
});
return { text, images: imageResult.images.map(toDocumentImage) };
const images: DocumentExtractedImage[] = [];
let remainingPixels = request.maxPixels;
for (let index = 0; index < imagePages.length; index += 1) {
if (remainingPixels <= 0) {
break;
}
const pagesRemaining = imagePages.length - index;
const maxPixelsPerPage = Math.max(1, Math.ceil(remainingPixels / pagesRemaining));
const pageNumber = imagePages[index];
const imageResult = await pdf.extract({
mode: "images",
pages: [pageNumber],
image: {
maxDimension: MAX_RENDER_DIMENSION,
maxPixels: maxPixelsPerPage,
forms: true,
},
});
for (const image of imageResult.images) {
images.push(toDocumentImage(image));
remainingPixels -= image.width * image.height;
}
}
return { text, images };
} catch (err) {
request.onImageExtractionError?.(err);
return { text, images: [] };