Datasets:

LeNcyer
/

SciFigBench

qid stringlengths 12 12	task stringclasses 1 value	sample_id stringlengths 7 7	image_path stringlengths 18 18	image_size listlengths 2 2	prompt stringclasses 1 value	answer listlengths 1 6	boxes_xyxy listlengths 1 6	colors listlengths 1 6	meta dict
27cce9bead57	dense_content_perception	F000001	images/F000001.jpg	[ 1347, 399 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "P_{N}", "[prompt]", "Encoder", "Domain Index K", "\\mathcal{L}_q" ]	[ [ 192, 151, 222, 170 ], [ 371, 144, 462, 169 ], [ 540, 158, 629, 182 ], [ 605, 240, 670, 277 ], [ 610, 304, 669, 358 ] ]	[ "red", "orange", "yellow", "green", "blue" ]	{ "node_ids": [ 6, 9, 4, 17, 24 ] }
68a1b39d69ee	dense_content_perception	F000002	images/F000002.jpg	[ 1515, 465 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Value" ]	[ [ 1063, 120, 1137, 153 ] ]	[ "red" ]	null
7eac00c4ab0c	dense_content_perception	F000002	images/F000002.jpg	[ 1515, 465 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "\"A 360 View of A Building\"", "Query", "Guidance" ]	[ [ 437, 10, 757, 40 ], [ 1061, 63, 1137, 102 ], [ 1445, 190, 1475, 288 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 0, 2, 10 ] }
a1bca5e06cd9	dense_content_perception	F000003	images/F000003.jpg	[ 1353, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Promising" ]	[ [ 985, 225, 1086, 252 ] ]	[ "red" ]	null
08915f84bc05	dense_content_perception	F000003	images/F000003.jpg	[ 1353, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "\\mathcal{D}", "\\lambda \\in \\Lambda_{random} \\quad P_{\\mathcal{D}_{val}}(A(\\mathcal{D}_{train}, \\lambda))", "\\mathcal{U} = \\{P_{i,1} \\succ P_{i,2}\\}_{i=1}^{U} \\quad u: \\mathcal{P} \\to \\mathbb{R}", "Get Pareto front P_{\\mathcal{D}_{val}}(A(\\mathcal{D}_{train}, \\lambda))" ]	[ [ 42, 43, 68, 66 ], [ 11, 270, 399, 298 ], [ 484, 319, 848, 349 ], [ 1065, 158, 1253, 219 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 0, 5, 4, 3 ] }
03fcf78c0b98	dense_content_perception	F000004	images/F000004.jpg	[ 1194, 537 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Model" ]	[ [ 639, 386, 751, 425 ] ]	[ "red" ]	null
e33a0407956d	dense_content_perception	F000004	images/F000004.jpg	[ 1194, 537 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "semantic guidance", "SEGA", "semantic guidance", "Model", "/transformed" ]	[ [ 599, 36, 792, 61 ], [ 640, 131, 749, 171 ], [ 599, 255, 792, 280 ], [ 639, 386, 751, 425 ], [ 985, 150, 1175, 188 ] ]	[ "red", "orange", "yellow", "green", "blue" ]	{ "node_ids": [ 0, 1, 4, 6, 2 ] }
9469a1fbf326	dense_content_perception	F000005	images/F000005.jpg	[ 717, 255 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "O_{[n],0}" ]	[ [ 330, 32, 377, 55 ] ]	[ "red" ]	null
29a1fee99f8b	dense_content_perception	F000005	images/F000005.jpg	[ 717, 255 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "O_{[n],0}", "a_{[n],0,f_0}", "\\left(O_{\\mathcal{I}_{0,0},1}\\right)", "t I_{0,0,1}" ]	[ [ 330, 32, 377, 55 ], [ 146, 117, 206, 139 ], [ 105, 210, 160, 233 ], [ 9, 192, 70, 244 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 0, 2, 4, 8 ] }
4989f68b5df1	dense_content_perception	F000006	images/F000006.jpg	[ 1323, 288 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Coding" ]	[ [ 794, 239, 851, 260 ] ]	[ "red" ]	null
3f809f1d1368	dense_content_perception	F000006	images/F000006.jpg	[ 1323, 288 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "What will Jerry do if he does", "Encoder", "Connected" ]	[ [ 29, 226, 237, 244 ], [ 523, 112, 588, 130 ], [ 654, 113, 725, 130 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 23, 13, 16 ] }
81c3e6d4d981	dense_content_perception	F000007	images/F000007.jpg	[ 1047, 405 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "cur" ]	[ [ 318, 129, 348, 144 ] ]	[ "red" ]	null
0e22434177f5	dense_content_perception	F000007	images/F000007.jpg	[ 1047, 405 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Previous", "Slope", "LastMax", "Count" ]	[ [ 119, 210, 200, 235 ], [ 382, 209, 440, 237 ], [ 615, 68, 693, 91 ], [ 866, 209, 926, 235 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 1, 2, 0, 3 ] }
d117ca39ff7d	dense_content_perception	F000008	images/F000008.jpg	[ 690, 270 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Agent2" ]	[ [ 598, 205, 668, 235 ] ]	[ "red" ]	null
14a2213533e1	dense_content_perception	F000008	images/F000008.jpg	[ 690, 270 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Agent1", "Agent_{0}", "Agent_{0}" ]	[ [ 599, 42, 668, 70 ], [ 343, 130, 414, 159 ], [ 599, 130, 667, 158 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 0, 3, 1 ] }
9fa9315f44d4	dense_content_perception	F000009	images/F000009.jpg	[ 603, 441 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "α" ]	[ [ 147, 305, 178, 327 ] ]	[ "red" ]	null
d4ba1674d57a	dense_content_perception	F000009	images/F000009.jpg	[ 603, 441 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Context", "R^{0}", "Agg.", "R^{M}", "\\pi", "a_{t}" ]	[ [ 217, 413, 318, 441 ], [ 327, 268, 360, 303 ], [ 403, 71, 461, 102 ], [ 529, 268, 568, 302 ], [ 46, 277, 74, 303 ], [ 49, 171, 78, 198 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 10, 6, 2, 7, 8, 5 ] }
dd92aef4903f	dense_content_perception	F000010	images/F000010.jpg	[ 720, 309 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Prompt" ]	[ [ 287, 41, 351, 62 ] ]	[ "red" ]	null
8cedb6318ae5	dense_content_perception	F000011	images/F000011.jpg	[ 720, 210 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Input" ]	[ [ 118, 55, 166, 78 ] ]	[ "red" ]	null
089bd1be7b6a	dense_content_perception	F000011	images/F000011.jpg	[ 720, 210 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Output", "Q-Connector", "UniGen", "D-Connector" ]	[ [ 344, 54, 406, 78 ], [ 215, 146, 300, 159 ], [ 357, 138, 429, 159 ], [ 509, 114, 594, 127 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 3, 9, 8, 6 ] }
b6a3aef0748d	dense_content_perception	F000012	images/F000012.jpg	[ 1395, 240 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "* W_3" ]	[ [ 929, 151, 1011, 196 ] ]	[ "red" ]	null
aa47cc4e7fc6	dense_content_perception	F000012	images/F000012.jpg	[ 1395, 240 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "observation", "reliability", "Policy" ]	[ [ 467, 22, 640, 59 ], [ 767, 117, 910, 154 ], [ 1039, 101, 1136, 137 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 1, 2, 3 ] }
84c9cbd70d57	dense_content_perception	F000013	images/F000013.jpg	[ 1479, 348 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "CNN feature" ]	[ [ 301, 280, 418, 305 ] ]	[ "red" ]	null
45a2263bc85e	dense_content_perception	F000013	images/F000013.jpg	[ 1479, 348 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "RoIAlign &", "Wij" ]	[ [ 603, 253, 702, 279 ], [ 831, 268, 872, 296 ] ]	[ "red", "orange" ]	{ "node_ids": [ 16, 18 ] }
7575823aeccc	dense_content_perception	F000014	images/F000014.jpg	[ 1185, 411 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "document" ]	[ [ 252, 169, 329, 187 ] ]	[ "red" ]	null
7004f99ce35a	dense_content_perception	F000014	images/F000014.jpg	[ 1185, 411 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "by 2025. Snickers bars", "document representation", "swin stage" ]	[ [ 8, 78, 162, 95 ], [ 443, 166, 625, 183 ], [ 1016, 20, 1100, 44 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 2, 0, 1 ] }
e4835a48d459	dense_content_perception	F000015	images/F000015.jpg	[ 1239, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "\\lambda_{2}" ]	[ [ 737, 193, 761, 214 ] ]	[ "red" ]	null
bf47f536f965	dense_content_perception	F000015	images/F000015.jpg	[ 1239, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Target", "Quantification", "Adaptive Tuning", "Augmentation", "Buffer", "Consistency Loss" ]	[ [ 102, 333, 158, 355 ], [ 248, 271, 424, 297 ], [ 501, 251, 711, 285 ], [ 518, 133, 696, 161 ], [ 303, 134, 370, 158 ], [ 770, 128, 999, 160 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 3, 6, 5, 1, 2, 0 ] }
0828cd15eb02	dense_content_perception	F000016	images/F000016.jpg	[ 1461, 426 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "A:[Write" ]	[ [ 911, 161, 976, 179 ] ]	[ "red" ]	null
e312b5505669	dense_content_perception	F000016	images/F000016.jpg	[ 1461, 426 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Seed Instructions", "Please comp up with 10 length-controlled text generation", "9: [Craft a text, approximately {number} words long:]", "Instructions", "Diversified", "{number} words: rewrite this" ]	[ [ 643, 96, 806, 116 ], [ 36, 98, 491, 120 ], [ 35, 372, 448, 397 ], [ 651, 170, 798, 196 ], [ 652, 301, 799, 329 ], [ 1134, 203, 1362, 222 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 2, 0, 7, 3, 6, 1 ] }
f9fded71461c	dense_content_perception	F000017	images/F000017.jpg	[ 744, 402 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Share" ]	[ [ 260, 153, 310, 177 ] ]	[ "red" ]	null
224916178912	dense_content_perception	F000017	images/F000017.jpg	[ 744, 402 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "DINOv2", "DINOv2", "\\mathbf{h}_{l}^{(i)}", "Concat", "layer", "Prediction" ]	[ [ 213, 89, 279, 108 ], [ 213, 241, 280, 263 ], [ 357, 237, 389, 267 ], [ 454, 240, 515, 263 ], [ 569, 249, 618, 277 ], [ 667, 231, 744, 252 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 2, 4, 5, 6, 7, 8 ] }
671aa1c1651b	dense_content_perception	F000018	images/F000018.jpg	[ 654, 699 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "60" ]	[ [ 246, 536, 277, 562 ] ]	[ "red" ]	null
2136b8ff971c	dense_content_perception	F000018	images/F000018.jpg	[ 654, 699 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Neural Network f_{\\theta}", "α", "(\\mu, \\sigma^2) \\sim \\frac{\\beta^{\\alpha} \\sqrt{v}}{\\Gamma(\\alpha) \\sqrt{2\\pi} \\sigma^2} \\left(\\frac{1}{\\sigma^2}\\right)^{\\alpha+1} \\exp\\left(-\\frac{2\\beta + v(\\gamma - \\mu)^2}{2\\sigma^2}\\right)", "y \\sim \\mathcal{N}(\\mu, \\sigma^{2}) \\Rightarro...	[ [ 76, 25, 115, 222 ], [ 244, 122, 265, 142 ], [ 321, 70, 640, 112 ], [ 334, 166, 625, 204 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 0, 9, 5, 10 ] }
e9beb68181e7	dense_content_perception	F000019	images/F000019.jpg	[ 1437, 507 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Mask k" ]	[ [ 1097, 101, 1153, 143 ] ]	[ "red" ]	null
f2bbcffca196	dense_content_perception	F000019	images/F000019.jpg	[ 1437, 507 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Features", "Features with Rel.", "Visible-Branch", "Amodal-Branch", "Coarse Amodal Mask I", "Output I" ]	[ [ 57, 272, 128, 319 ], [ 368, 271, 442, 334 ], [ 631, 14, 765, 30 ], [ 631, 196, 776, 212 ], [ 912, 237, 1033, 323 ], [ 1339, 187, 1402, 232 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 7, 4, 0, 5, 6, 3 ] }
ee2d375c277d	dense_content_perception	F000020	images/F000020.jpg	[ 603, 555 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Encoder" ]	[ [ 294, 424, 392, 453 ] ]	[ "red" ]	null
31c074a3c4ca	dense_content_perception	F000020	images/F000020.jpg	[ 603, 555 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Backbone", "sion Module", "Head" ]	[ [ 130, 23, 226, 46 ], [ 256, 84, 372, 108 ], [ 406, 71, 461, 97 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 1, 3, 4 ] }
bfdb335432d7	dense_content_perception	F000021	images/F000021.jpg	[ 1053, 462 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Encoder" ]	[ [ 52, 294, 138, 314 ] ]	[ "red" ]	null
fe769fe8fef2	dense_content_perception	F000021	images/F000021.jpg	[ 1053, 462 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "During Training", "Adaptor", "Language Model" ]	[ [ 340, 23, 491, 49 ], [ 605, 249, 693, 277 ], [ 826, 247, 1001, 274 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 0, 8, 7 ] }
799c6cd6d204	dense_content_perception	F000022	images/F000022.jpg	[ 711, 444 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "HopGraph" ]	[ [ 406, 396, 493, 419 ] ]	[ "red" ]	null
5ab120917ea0	dense_content_perception	F000022	images/F000022.jpg	[ 711, 444 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "HopGraph", "distance matrix", "hop distance loss" ]	[ [ 406, 396, 493, 419 ], [ 387, 287, 513, 303 ], [ 567, 192, 703, 214 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 16, 13, 7 ] }
e5eeaf68d70f	dense_content_perception	F000023	images/F000023.jpg	[ 1374, 564 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "data" ]	[ [ 172, 465, 223, 493 ] ]	[ "red" ]	null
e32ff3f967a6	dense_content_perception	F000023	images/F000023.jpg	[ 1374, 564 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "image/video", "data as input", "\\sim Q" ]	[ [ 24, 432, 150, 458 ], [ 1008, 245, 1141, 274 ], [ 1319, 434, 1362, 464 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 7, 0, 5 ] }
756120c42906	dense_content_perception	F000024	images/F000024.jpg	[ 732, 525 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Sampling" ]	[ [ 116, 250, 189, 273 ] ]	[ "red" ]	null
bc9eff6a0b20	dense_content_perception	F000024	images/F000024.jpg	[ 732, 525 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Sampling", "Dense embedding", "Surface point cloud" ]	[ [ 116, 250, 189, 273 ], [ 364, 32, 501, 54 ], [ 204, 311, 354, 328 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 1, 0, 5 ] }
9e3f47353cac	dense_content_perception	F000025	images/F000025.jpg	[ 684, 471 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Attention" ]	[ [ 550, 355, 621, 373 ] ]	[ "red" ]	null
74736ffaa829	dense_content_perception	F000025	images/F000025.jpg	[ 684, 471 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "carrying a wounded Syrian", "Text encoder", "Gated attention", "Multi-Head", "Add & Norm" ]	[ [ 6, 86, 193, 106 ], [ 237, 75, 332, 93 ], [ 385, 76, 497, 92 ], [ 543, 196, 624, 212 ], [ 538, 150, 630, 168 ] ]	[ "red", "orange", "yellow", "green", "blue" ]	{ "node_ids": [ 1, 4, 3, 9, 6 ] }
6d5d0cf494e6	dense_content_perception	F000026	images/F000026.jpg	[ 717, 468 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Z" ]	[ [ 397, 67, 422, 94 ] ]	[ "red" ]	null
e576301fe3df	dense_content_perception	F000026	images/F000026.jpg	[ 717, 468 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "N", "\\widetilde{G}" ]	[ [ 257, 182, 277, 202 ], [ 695, 59, 717, 93 ] ]	[ "red", "orange" ]	{ "node_ids": [ 1, 4 ] }
11152f753ecc	dense_content_perception	F000027	images/F000027.jpg	[ 1500, 396 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Model" ]	[ [ 1370, 241, 1440, 262 ] ]	[ "red" ]	null
885fa218b8b8	dense_content_perception	F000027	images/F000027.jpg	[ 1500, 396 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Pruning", "Estimation", "Verification" ]	[ [ 1044, 255, 1139, 279 ], [ 1345, 271, 1465, 293 ], [ 1340, 117, 1471, 139 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 15, 14, 3 ] }
11a8d5390aeb	dense_content_perception	F000028	images/F000028.jpg	[ 1101, 531 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Q K V" ]	[ [ 624, 266, 730, 294 ] ]	[ "red" ]	null
698839857c3a	dense_content_perception	F000028	images/F000028.jpg	[ 1101, 531 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Input Text", "Corrupt & Shuffle", "Text Embedding" ]	[ [ 403, 401, 497, 428 ], [ 371, 346, 528, 371 ], [ 375, 281, 521, 307 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 13, 11, 8 ] }
ac463820a64d	dense_content_perception	F000029	images/F000029.jpg	[ 1296, 621 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Do you know [DEL]?" ]	[ [ 548, 437, 699, 457 ] ]	[ "red" ]	null
c5f06f5c7bbb	dense_content_perception	F000029	images/F000029.jpg	[ 1296, 621 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "\\langle C^{-},K^{-}\\rangle\\ \\langle C^{+},K^{+}\\rangle\\ \\langle C,K\\rangle", "Context-Knowledge Fusion", "negative", "\\mathcal{L}_{cl}" ]	[ [ 964, 401, 1165, 425 ], [ 982, 289, 1141, 306 ], [ 978, 193, 1022, 208 ], [ 1052, 118, 1096, 151 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 17, 9, 3, 2 ] }
55717c6f59fa	dense_content_perception	F000030	images/F000030.jpg	[ 1464, 372 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Requests" ]	[ [ 164, 293, 249, 316 ] ]	[ "red" ]	null
8429cc29c95b	dense_content_perception	F000030	images/F000030.jpg	[ 1464, 372 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Trajectories", "Modeling", "Simulator", "B", "A", "A" ]	[ [ 41, 95, 147, 117 ], [ 267, 91, 352, 114 ], [ 525, 123, 554, 241 ], [ 1349, 173, 1360, 187 ], [ 1170, 173, 1195, 197 ], [ 713, 122, 738, 148 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 1, 2, 0, 3, 7, 4 ] }
a63204dc698c	dense_content_perception	F000031	images/F000031.jpg	[ 687, 516 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "2000" ]	[ [ 577, 481, 612, 499 ] ]	[ "red" ]	null
0e22747babae	dense_content_perception	F000032	images/F000032.jpg	[ 1215, 513 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Extractor" ]	[ [ 500, 440, 608, 464 ] ]	[ "red" ]	null
8bf8a2414fc7	dense_content_perception	F000032	images/F000032.jpg	[ 1215, 513 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "alignment exceptions", "Log-Attention" ]	[ [ 253, 157, 471, 187 ], [ 809, 119, 964, 147 ] ]	[ "red", "orange" ]	{ "node_ids": [ 0, 3 ] }
28a661642670	dense_content_perception	F000033	images/F000033.jpg	[ 1428, 234 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Detection" ]	[ [ 349, 121, 436, 136 ] ]	[ "red" ]	null
3779fefcb46d	dense_content_perception	F000033	images/F000033.jpg	[ 1428, 234 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "trained", "Detection", "Query Samples", "Classifier Model", "Prediction", "Inconsistency" ]	[ [ 241, 95, 303, 112 ], [ 349, 121, 436, 136 ], [ 495, 93, 630, 114 ], [ 681, 170, 822, 188 ], [ 864, 171, 961, 188 ], [ 957, 52, 1081, 75 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 6, 7, 5, 8, 9, 4 ] }
0b1db55055a4	dense_content_perception	F000034	images/F000034.jpg	[ 1134, 390 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Logits" ]	[ [ 205, 96, 279, 123 ] ]	[ "red" ]	null
11d64c0f28b6	dense_content_perception	F000034	images/F000034.jpg	[ 1134, 390 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "\\{X_{L},Y_{L},X_{U}\\}", "Proposal", "Class-prob.", "Weak Scene", "Strong Scene", "Student" ]	[ [ 17, 200, 116, 227 ], [ 190, 307, 295, 331 ], [ 173, 185, 309, 211 ], [ 409, 307, 555, 329 ], [ 401, 69, 565, 96 ], [ 650, 70, 746, 92 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 7, 11, 6, 10, 2, 1 ] }
6907123f91ae	dense_content_perception	F000035	images/F000035.jpg	[ 1476, 588 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "O" ]	[ [ 614, 20, 644, 48 ] ]	[ "red" ]	null
d12f97b8afc3	dense_content_perception	F000061	images/F000061.jpg	[ 1248, 597 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "1", "Vision Transformer", "Terminate?", "discriminative Region", "7", "3" ]	[ [ 391, 36, 407, 57 ], [ 639, 192, 662, 398 ], [ 878, 89, 975, 110 ], [ 825, 220, 1035, 249 ], [ 761, 291, 778, 312 ], [ 414, 372, 424, 431 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 2, 1, 6, 8, 7, 11 ] }
1de3d0b6daf5	dense_content_perception	F000036	images/F000036.jpg	[ 534, 567 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Training" ]	[ [ 2, 101, 36, 198 ] ]	[ "red" ]	null
72c72e489db7	dense_content_perception	F000037	images/F000037.jpg	[ 711, 363 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "feature" ]	[ [ 237, 276, 300, 296 ] ]	[ "red" ]	null
4437ed89993e	dense_content_perception	F000037	images/F000037.jpg	[ 711, 363 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "(x, y, z, d)", "decoder", "Volume", "\\mathcal{L}_{\\text{content}}" ]	[ [ 36, 45, 112, 67 ], [ 260, 218, 324, 236 ], [ 398, 158, 460, 178 ], [ 622, 114, 711, 151 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 1, 7, 3, 4 ] }
ecd492697371	dense_content_perception	F000038	images/F000038.jpg	[ 723, 390 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "making" ]	[ [ 308, 78, 365, 98 ] ]	[ "red" ]	null
86da594724e6	dense_content_perception	F000038	images/F000038.jpg	[ 723, 390 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "feature n", "prediction: 1", "0" ]	[ [ 0, 209, 77, 229 ], [ 434, 142, 516, 161 ], [ 578, 195, 586, 203 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 2, 1, 0 ] }
72488cfbca32	dense_content_perception	F000039	images/F000039.jpg	[ 660, 444 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Policy" ]	[ [ 175, 226, 256, 264 ] ]	[ "red" ]	null
8d525613df94	dense_content_perception	F000039	images/F000039.jpg	[ 660, 444 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Datasets", "Dynamics Model", "Rollout" ]	[ [ 102, 14, 203, 45 ], [ 14, 99, 212, 132 ], [ 42, 227, 136, 260 ] ]	[ "red", "orange", "yellow" ]	{ "node_ids": [ 0, 3, 4 ] }
767c67a52178	dense_content_perception	F000040	images/F000040.jpg	[ 1506, 303 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Graph" ]	[ [ 276, 185, 359, 218 ] ]	[ "red" ]	null
9c159440fa42	dense_content_perception	F000040	images/F000040.jpg	[ 1506, 303 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "A Test Graph", "Graph", "Masked Graph", "Trained", "Masked Graph", "GIB-boosted Losses" ]	[ [ 42, 199, 189, 226 ], [ 276, 185, 359, 218 ], [ 535, 209, 698, 232 ], [ 748, 117, 832, 138 ], [ 891, 178, 1054, 205 ], [ 1120, 210, 1373, 239 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 4, 2, 1, 6, 3, 0 ] }
a2041b643287	dense_content_perception	F000041	images/F000041.jpg	[ 1152, 471 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "-2" ]	[ [ 615, 128, 645, 157 ] ]	[ "red" ]	null
88dd6a2e95d0	dense_content_perception	F000042	images/F000042.jpg	[ 717, 564 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Answer:..." ]	[ [ 290, 205, 422, 243 ] ]	[ "red" ]	null
7007ac4c833e	dense_content_perception	F000042	images/F000042.jpg	[ 717, 564 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Question: An animal might pant_", "Answer:..." ]	[ [ 262, 9, 637, 41 ], [ 290, 205, 422, 243 ] ]	[ "red", "orange" ]	{ "node_ids": [ 0, 2 ] }
d311162c01d2	dense_content_perception	F000043	images/F000043.jpg	[ 1473, 555 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Input I" ]	[ [ 40, 71, 130, 103 ] ]	[ "red" ]	null
ae684cf65459	dense_content_perception	F000043	images/F000043.jpg	[ 1473, 555 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Attention", "Inpainting" ]	[ [ 1004, 315, 1110, 340 ], [ 1325, 264, 1450, 297 ] ]	[ "red", "orange" ]	{ "node_ids": [ 12, 11 ] }
0ca3cb539589	dense_content_perception	F000044	images/F000044.jpg	[ 1392, 666 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Input" ]	[ [ 39, 134, 109, 168 ] ]	[ "red" ]	null
a24c299f4eea	dense_content_perception	F000044	images/F000044.jpg	[ 1392, 666 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Background Affine Transformation", "Dense Motion", "Dense Motion", "Inpainting", "DINO ViT", "\\mathcal{L}_{\\text{struc}}" ]	[ [ 362, 48, 689, 70 ], [ 847, 101, 969, 123 ], [ 1141, 142, 1275, 164 ], [ 1036, 265, 1127, 294 ], [ 1201, 415, 1290, 434 ], [ 1205, 489, 1283, 533 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 2, 4, 1, 6, 11, 16 ] }
dcef4814e707	dense_content_perception	F000045	images/F000045.jpg	[ 1338, 528 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Layer 2" ]	[ [ 284, 186, 350, 212 ] ]	[ "red" ]	null
3108a7b6743e	dense_content_perception	F000045	images/F000045.jpg	[ 1338, 528 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Layer L - 1", "Layer L-1" ]	[ [ 935, 141, 1031, 163 ], [ 1162, 142, 1255, 163 ] ]	[ "red", "orange" ]	{ "node_ids": [ 1, 2 ] }
1549e41307ad	dense_content_perception	F000046	images/F000046.jpg	[ 726, 702 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Interaction" ]	[ [ 549, 678, 649, 697 ] ]	[ "red" ]	null
7543870b8e73	dense_content_perception	F000046	images/F000046.jpg	[ 726, 702 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Action:", "Symbolic", "same quantity as the answer, and place", "LLM Agent", "mul 26 36" ]	[ [ 500, 211, 563, 234 ], [ 622, 248, 691, 269 ], [ 14, 423, 293, 443 ], [ 340, 440, 422, 460 ], [ 494, 413, 570, 433 ] ]	[ "red", "orange", "yellow", "green", "blue" ]	{ "node_ids": [ 3, 1, 7, 8, 5 ] }
6ef9b3b56b28	dense_content_perception	F000047	images/F000047.jpg	[ 1350, 771 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Backbone" ]	[ [ 256, 56, 398, 90 ] ]	[ "red" ]	null
76593662d3bc	dense_content_perception	F000047	images/F000047.jpg	[ 1350, 771 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Layer index", "Layer-wise Penalty", "\\hat{\\theta}_{t+1} - \\lambda_i(\\hat{\\theta}_{t+1} - \\theta_0)", "Upstream data" ]	[ [ 1064, 696, 1234, 730 ], [ 1016, 565, 1279, 599 ], [ 997, 428, 1299, 478 ], [ 1084, 71, 1291, 109 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 14, 11, 10, 1 ] }
f6441e133fd4	dense_content_perception	F000048	images/F000048.jpg	[ 927, 525 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Subobject" ]	[ [ 728, 162, 805, 185 ] ]	[ "red" ]	null
89cff8848e9a	dense_content_perception	F000048	images/F000048.jpg	[ 927, 525 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "offensive gestures", "X", "Conditional Image", "Large Language Model" ]	[ [ 13, 235, 106, 251 ], [ 147, 227, 167, 246 ], [ 198, 197, 332, 216 ], [ 179, 118, 345, 136 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 7, 9, 6, 1 ] }
5266309e5d52	dense_content_perception	F000049	images/F000049.jpg	[ 711, 276 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Phase 1" ]	[ [ 47, 102, 108, 119 ] ]	[ "red" ]	null
62067f731755	dense_content_perception	F000049	images/F000049.jpg	[ 711, 276 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Data(S, X^1)", "Generator", "Generated", "\\mathrm{Dist}\\left(X_+^T, X_-^T\\right)", "\\mathbf{h}_{\\theta}", "\\mathbf{h}_{\\theta}" ]	[ [ 509, 54, 601, 76 ], [ 515, 117, 586, 135 ], [ 511, 182, 583, 198 ], [ 505, 225, 609, 251 ], [ 649, 205, 671, 224 ], [ 632, 117, 654, 135 ] ]	[ "red", "orange", "yellow", "green", "blue", "violet" ]	{ "node_ids": [ 1, 4, 11, 13, 12, 7 ] }
77eed4bd938b	dense_content_perception	F000050	images/F000050.jpg	[ 1413, 717 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Consistency" ]	[ [ 932, 131, 1048, 163 ] ]	[ "red" ]	null
d5ef1c218441	dense_content_perception	F000050	images/F000050.jpg	[ 1413, 717 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Input Point Cloud Sequence", "Prediction", "Predicted BEV Motion", "Consistency" ]	[ [ 5, 16, 255, 44 ], [ 391, 96, 489, 124 ], [ 582, 19, 777, 40 ], [ 932, 131, 1048, 163 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 0, 2, 1, 3 ] }
b7797ae090db	dense_content_perception	F000051	images/F000051.jpg	[ 1266, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Encoder" ]	[ [ 462, 224, 493, 327 ] ]	[ "red" ]	null
72c0414042a2	dense_content_perception	F000051	images/F000051.jpg	[ 1266, 384 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Images", "Feature", "P_{\|\\tilde{c}\|}", "\\mathcal{L}_{\\mathrm{DW}}^{\\mathrm{P}}" ]	[ [ 226, 233, 297, 261 ], [ 430, 230, 463, 327 ], [ 518, 121, 552, 147 ], [ 919, 104, 973, 137 ] ]	[ "red", "orange", "yellow", "green" ]	{ "node_ids": [ 11, 10, 0, 5 ] }
7deb2845838f	dense_content_perception	F000052	images/F000052.jpg	[ 1509, 813 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "0" ]	[ [ 1055, 797, 1071, 812 ] ]	[ "red" ]	null
1dbc248f9078	dense_content_perception	F000052	images/F000052.jpg	[ 1509, 813 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Logit Layer", "Encoder" ]	[ [ 776, 436, 915, 471 ], [ 791, 586, 892, 612 ] ]	[ "red", "orange" ]	{ "node_ids": [ 10, 12 ] }
a7adb5effe52	dense_content_perception	F000053	images/F000053.jpg	[ 1212, 438 ]	Recognize the text inside each colored bounding box. Process boxes in this fixed color order: red → orange → yellow → green → blue → violet. Return a JSON list of strings, with one string for each present bounding box. Skip missing colors and do not output empty strings for them. The list length K must equal the numbe...	[ "Inv Obj" ]	[ [ 1092, 101, 1165, 129 ] ]	[ "red" ]	null

End of preview. Expand in Data Studio

SciFigBench: A Benchmark for Scientific Method Figure Understanding

SciFigBench is a diagnostic benchmark for scientific method figure understanding, introduced in our EMNLP 2026 Main Conference paper. It comprises 4,352 real-world scientific method figures and 40,670 QA pairs across five tasks and three core capabilities: Dense Content Perception, Visual Relation Comprehension, and Information-Flow Tracing.

The dataset also provides content and structural annotations, which underwent a second round of human verification.

Resources: Paper | GitHub | Annotation Schema

Dataset Overview

Left: figure sources across seven AI conferences. Right: QA distribution across five tasks.

Dataset Structure

SciFigBench provides five task-specific QA files, 4,352 images, and 4,352 annotation files.

SciFigBench/
├── QA/                # Five task-specific JSONL files
│   ├── manifest.json
│   └── checksums.sha256
├── images/            # 4,352 scientific method figures
├── annotations/       # 4,352 annotation JSON files
├── assets/            # Dataset Card illustrations
├── README.md
├── LICENSE
├── manifest.jsonl
├── schema.md
└── checksums.sha256

All resources are linked through anonymized sample_id identifiers (F000001–F004352). Each annotation contains content_annotation (text fields) and structural_annotation (nodes and arrow endpoints).

See schema.md for detailed field definitions and coordinate conventions.

Usage

Download the QA files and images from the latest public version using Hugging Face Hub:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="LeNcyer/SciFigBench",
    repo_type="dataset",
    revision="main",  # Latest public version; this branch can change.
    local_dir="./data",
    allow_patterns=["QA/*", "images/*.jpg", "checksums.sha256"],
)

For a reproducible accepted-data snapshot, replace revision="main" with revision="0e7244ab8aef016db34fe0c8fc0598a5eac2f7fe"; main is mutable. QA/* and images/*.jpg supply the standard evaluation inputs; the checksum files verify their integrity. Optional "annotations/*.json", "schema.md", and "manifest.jsonl" can be added to allow_patterns for annotation research and schema/reference information.

For standard VLM evaluation, use only figure images, task prompts, and query-specific highlights. Ground-truth annotations should not be provided as additional inference-time inputs.

The official evaluation toolkit and usage instructions are available on GitHub.

License & Citation

The SciFigBench dataset is released under the MIT License. See LICENSE.

If you find our work helpful, please cite:

@inproceedings{wu2026scifigbench,
  title     = {SciFigBench: A Benchmark for Scientific Method Figure Understanding},
  author    = {Wu, Lijia and Li, Zheng and Liu, Yuxuan and Zhang, Jingwen and Zhao, Penghai and Li, Xiang},
  booktitle = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing},
  year      = {2026},
  eprint    = {2610.27087},
  archivePrefix = {arXiv},
  url       = {https://arxiv.org/abs/2610.27087}
}

Downloads last month: 96

Paper for LeNcyer/SciFigBench

SciFigBench: A Benchmark for Scientific Method Figure Understanding

Paper • 2610.27087 • Published 3 days ago