Datasets:

gamusa
/

docvqa-canonical

sample_id stringlengths 10 12	domain_type stringclasses 1 value	split stringclasses 1 value	question_id stringlengths 3 5	question stringlengths 11 168	question_tokens listlengths 2 30	image_id stringlengths 14 16	image dict	image_width int64 200 7.07k	image_height int64 238 9.68k	answers listlengths 1 14	canonical_answer stringlengths 1 187	ocr_tokens listlengths 0 1.84k	ocr_boxes listlengths 0 1.84k	metadata stringlengths 104 142
docvqa_337	document_text	train	337	what is the date mentioned in this letter?	[ "what", "is", "the", "date", "mentioned", "in", "this", "letter?" ]	xnbl0037_1.png	{ "bytes": [ 137, 80, 78, 71, 13, 10, 26, 10, 0, 0, 0, 13, 73, 72, 68, 82, 0, 0, 6, 159, 0, 0, 7, 233, 8, 0, 0, 0, 0, 50, 25, 78, 45, 0, 0, 224, 97, 73, 68, 6...	1,695	2,025	[ "1/8/93" ]	1/8/93	[ "Confidential", "..", "..", "RJRT", "PR", "APPROVAL", "DATE", ":", "1/8/13", "Ru", "alAs", "PROPOSED", "RELEASE", "DATE:", "for", "response", "FOR", "RELEASE", "TO:", "CONTACT:", "P.", "CARTER", "ROUTE", "TO", "Initials", "pate", "Peggy", "Carter", "Ac", "Ma...	[ [ 15, 345, 65, 586 ], [ 16, 637, 27, 648 ], [ 15, 650, 27, 663 ], [ 100, 412, 116, 460 ], [ 101, 468, 116, 491 ], [ 101, 500, 117, 587 ], [ 146, 153, 160, 195 ], [ 146, ...	{"question_types": ["handwritten", "form"], "docId": 279, "ucsf_document_id": "xnbl0037", "ucsf_document_page_no": "1"}
docvqa_338	document_text	train	338	what is the contact person name mentioned in letter?	[ "what", "is", "the", "contact", "person", "name", "mentioned", "in", "letter?" ]	xnbl0037_1.png	{ "bytes": [ 137, 80, 78, 71, 13, 10, 26, 10, 0, 0, 0, 13, 73, 72, 68, 82, 0, 0, 6, 159, 0, 0, 7, 233, 8, 0, 0, 0, 0, 50, 25, 78, 45, 0, 0, 224, 97, 73, 68, 6...	1,695	2,025	[ "P. Carter", "p. carter" ]	P. Carter	[ "Confidential", "..", "..", "RJRT", "PR", "APPROVAL", "DATE", ":", "1/8/13", "Ru", "alAs", "PROPOSED", "RELEASE", "DATE:", "for", "response", "FOR", "RELEASE", "TO:", "CONTACT:", "P.", "CARTER", "ROUTE", "TO", "Initials", "pate", "Peggy", "Carter", "Ac", "Ma...	[ [ 15, 345, 65, 586 ], [ 16, 637, 27, 648 ], [ 15, 650, 27, 663 ], [ 100, 412, 116, 460 ], [ 101, 468, 116, 491 ], [ 101, 500, 117, 587 ], [ 146, 153, 160, 195 ], [ 146, ...	{"question_types": ["handwritten", "form"], "docId": 279, "ucsf_document_id": "xnbl0037", "ucsf_document_page_no": "1"}
docvqa_339	document_text	train	339	Which corporation's letterhead is this?	[ "which", "corporation's", "letterhead", "is", "this?" ]	mxcj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABkYAAAg4CAAAAAC+N1XNAAEAAElEQVR4nOzdd3gUVcPG4V/oPQmQ0EnonawFbEg2CF(...TRUNCATED)	1,606	2,104	[ "Brown & Williamson Tobacco Corporation" ]	Brown & Williamson Tobacco Corporation	["B&W","BROWN","&","WILLIAMSON","TOBACCO","CORPORATION","RESEARCH","&","DEVELOPMENT",".",".",".","."(...TRUNCATED)	[[92,456,104,493],[119,297,131,353],[119,356,131,369],[119,373,130,471],[119,476,130,549],[118,553,1(...TRUNCATED)	"{\"question_types\": [\"layout\"], \"docId\": 280, \"ucsf_document_id\": \"mxcj0037\", \"ucsf_docum(...TRUNCATED)
docvqa_340	document_text	train	340	Who is in cc in this letter?	[ "who", "is", "in", "cc", "in", "this", "letter?" ]	mxcj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABkYAAAg4CAAAAAC+N1XNAAEAAElEQVR4nOzdd3gUVcPG4V/oPQmQ0EnonawFbEg2CF(...TRUNCATED)	1,606	2,104	[ "T.F. Riehl" ]	T.F. Riehl	["B&W","BROWN","&","WILLIAMSON","TOBACCO","CORPORATION","RESEARCH","&","DEVELOPMENT",".",".",".","."(...TRUNCATED)	[[92,456,104,493],[119,297,131,353],[119,356,131,369],[119,373,130,471],[119,476,130,549],[118,553,1(...TRUNCATED)	"{\"question_types\": [\"form\", \"layout\"], \"docId\": 280, \"ucsf_document_id\": \"mxcj0037\", \"(...TRUNCATED)
docvqa_341	document_text	train	341	what is the subject of this letter?	[ "what", "is", "the", "subject", "of", "this", "letter?" ]	mxcj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABkYAAAg4CAAAAAC+N1XNAAEAAElEQVR4nOzdd3gUVcPG4V/oPQmQ0EnonawFbEg2CF(...TRUNCATED)	1,606	2,104	[ "Review of existing Brainstorming Ideas/483" ]	Review of existing Brainstorming Ideas/483	["B&W","BROWN","&","WILLIAMSON","TOBACCO","CORPORATION","RESEARCH","&","DEVELOPMENT",".",".",".","."(...TRUNCATED)	[[92,456,104,493],[119,297,131,353],[119,356,131,369],[119,373,130,471],[119,476,130,549],[118,553,1(...TRUNCATED)	"{\"question_types\": [\"form\", \"layout\"], \"docId\": 280, \"ucsf_document_id\": \"mxcj0037\", \"(...TRUNCATED)
docvqa_343	document_text	train	343	What is the number at the bottom of the page, in bold?	[ "what", "is", "the", "number", "at", "the", "bottom", "of", "the", "page,", "in", "bold?" ]	mxcj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABkYAAAg4CAAAAAC+N1XNAAEAAElEQVR4nOzdd3gUVcPG4V/oPQmQ0EnonawFbEg2CF(...TRUNCATED)	1,606	2,104	[ "499150498" ]	499150498	["B&W","BROWN","&","WILLIAMSON","TOBACCO","CORPORATION","RESEARCH","&","DEVELOPMENT",".",".",".","."(...TRUNCATED)	[[92,456,104,493],[119,297,131,353],[119,356,131,369],[119,373,130,471],[119,476,130,549],[118,553,1(...TRUNCATED)	"{\"question_types\": [\"layout\"], \"docId\": 280, \"ucsf_document_id\": \"mxcj0037\", \"ucsf_docum(...TRUNCATED)
docvqa_388	document_text	train	388	How many points are there in modifications to readout instrumentation	[ "how", "many", "points", "are", "there", "in", "modifications", "to", "readout", "instrumentation" ]	sxxj0037_2.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABpwAAAjFCAAAAABfCECKAAEAAElEQVR4nOyda5rrqg6mpX7OvM7ukZmM7OwemfqHuU(...TRUNCATED)	1,692	2,245	[ "5.", "5" ]	5.	["2.","The","rotometer","in","the","original","design","was","replaced","by","a","0-100","SILPM","ma(...TRUNCATED)	[[135,202,147,215],[135,245,150,272],[135,276,150,345],[135,349,150,366],[135,369,150,393],[135,397,(...TRUNCATED)	"{\"question_types\": [\"table/list\"], \"docId\": 297, \"ucsf_document_id\": \"sxxj0037\", \"ucsf_d(...TRUNCATED)
docvqa_399	document_text	train	399	what is the date in the letter	[ "what", "is", "the", "date", "in", "the", "letter" ]	nhxj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABqoAAAiWCAAAAACw/8isAAEAAElEQVR4nOyda5azrBKFt2d985IeWeHIJCPj/OBWXD(...TRUNCATED)	1,706	2,198	[ "June 11, 1990", "June 11,1990" ]	June 11, 1990	["PHILIP","MORRIS","U.","S.","A.","INTER-OFFICE","CORRESPONDENCE","Richmond,","Virginia","To","Dr.",(...TRUNCATED)	[[81,348,96,410],[81,413,96,493],[81,500,96,518],[81,521,96,539],[81,542,96,559],[108,231,124,416],[(...TRUNCATED)	"{\"question_types\": [\"form\", \"layout\"], \"docId\": 314, \"ucsf_document_id\": \"nhxj0037\", \"(...TRUNCATED)
docvqa_400	document_text	train	400	To whom this is addressed	[ "to", "whom", "this", "is", "addressed" ]	nhxj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABqoAAAiWCAAAAACw/8isAAEAAElEQVR4nOyda5azrBKFt2d985IeWeHIJCPj/OBWXD(...TRUNCATED)	1,706	2,198	[ "Dr.K.S.Houghton" ]	Dr.K.S.Houghton	["PHILIP","MORRIS","U.","S.","A.","INTER-OFFICE","CORRESPONDENCE","Richmond,","Virginia","To","Dr.",(...TRUNCATED)	[[81,348,96,410],[81,413,96,493],[81,500,96,518],[81,521,96,539],[81,542,96,559],[108,231,124,416],[(...TRUNCATED)	"{\"question_types\": [\"form\", \"layout\"], \"docId\": 314, \"ucsf_document_id\": \"nhxj0037\", \"(...TRUNCATED)
docvqa_401	document_text	train	401	Who sent the letter?	[ "who", "sent", "the", "letter?" ]	nhxj0037_1.png	{"bytes":"iVBORw0KGgoAAAANSUhEUgAABqoAAAiWCAAAAACw/8isAAEAAElEQVR4nOyda5azrBKFt2d985IeWeHIJCPj/OBWXD(...TRUNCATED)	1,706	2,198	[ "Ted Sanders" ]	Ted Sanders	["PHILIP","MORRIS","U.","S.","A.","INTER-OFFICE","CORRESPONDENCE","Richmond,","Virginia","To","Dr.",(...TRUNCATED)	[[81,348,96,410],[81,413,96,493],[81,500,96,518],[81,521,96,539],[81,542,96,559],[108,231,124,416],[(...TRUNCATED)	"{\"question_types\": [\"form\", \"layout\"], \"docId\": 314, \"ucsf_document_id\": \"nhxj0037\", \"(...TRUNCATED)

End of preview. Expand in Data Studio

docvqa-canonical (Lossless Sharded Canonical)

Normalized Canonical Dataset for CrossVQA Benchmark (lossless branch).

Format preservation: 100% original image formats (Lossless PNG for DocVQA, Native JPEG for TextVQA).
Streaming Shards: Optimized with page index (write_page_index=True) for instant browsing on Hugging Face Dataset Viewer without timeouts.
Spatial & EXIF Alignment: All OCR boxes scaled to [ymin, xmin, ymax, xmax] in [0, 1000].

Note on Benchmark Splits (main branch): This repository contains curated train and validation splits for reproducible cross-domain VQA research. The blind challenge test split (which has withheld ground-truth labels) has been removed from this branch. For the complete unpruned benchmark, refer to the lossless branch.

Downloads last month: 110