%PDF-1.3 1 0 obj << /Kids [ 3 0 R 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R ] /Type /Pages /Count 9 >> endobj 2 0 obj << /Title (Unsupervised Learning and Segmentation of Complex Activities From Video) /Producer (PyPDF2) /Author (Fadime Sener\054 Angela Yao) /Subject (2018 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 13 0 R /Resources << /XObject << /x8 14 0 R /x6 17 0 R /x12 20 0 R /x10 23 0 R >> /ExtGState << /s9 26 0 R /R28 29 0 R /s11 30 0 R /a0 << /CA 1 /ca 1 >> /s5 33 0 R /s7 36 0 R >> /Font << /F2 39 0 R /R29 40 0 R /F1 43 0 R /R37 44 0 R /R35 48 0 R /R33 51 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 59 0 R 60 0 R 61 0 R 62 0 R 63 0 R 64 0 R 65 0 R 66 0 R 67 0 R 68 0 R 69 0 R 70 0 R 71 0 R 72 0 R 73 0 R 74 0 R 75 0 R 76 0 R 77 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 78 0 R /Resources << /XObject << /R75 79 0 R >> /ExtGState << /R28 29 0 R >> /Font << /R84 80 0 R /F2 83 0 R /R78 84 0 R /R73 88 0 R /F1 92 0 R /R76 93 0 R /R29 40 0 R /R88 97 0 R /R37 44 0 R /R90 101 0 R /R82 105 0 R /R80 109 0 R /R86 112 0 R /R31 55 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 117 0 R 118 0 R 119 0 R 120 0 R 121 0 R 122 0 R 123 0 R 124 0 R 125 0 R 126 0 R 127 0 R 128 0 R 129 0 R 130 0 R 131 0 R 132 0 R 133 0 R 134 0 R 135 0 R 136 0 R 137 0 R 138 0 R 139 0 R 140 0 R 141 0 R 142 0 R 143 0 R 144 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 145 0 R /Resources << /ExtGState << /R28 29 0 R >> /Font << /F1 146 0 R /R76 93 0 R /R127 147 0 R /R131 150 0 R /R125 153 0 R /R88 97 0 R /R123 157 0 R /R135 161 0 R /R121 165 0 R /R137 169 0 R /R90 101 0 R /R82 105 0 R /R80 109 0 R /R86 112 0 R /R129 173 0 R /R84 80 0 R /F2 177 0 R /R119 178 0 R /R141 182 0 R /R29 40 0 R /R133 185 0 R /R37 44 0 R /R139 188 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 192 0 R 193 0 R 194 0 R 195 0 R 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R 207 0 R 208 0 R 209 0 R 210 0 R 211 0 R 212 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 213 0 R /Resources << /ExtGState << /R28 29 0 R >> /Font << /F1 214 0 R /R76 93 0 R /R127 147 0 R /R125 153 0 R /R133 185 0 R /R123 157 0 R /R135 161 0 R /R121 165 0 R /R90 101 0 R /R139 188 0 R /R86 112 0 R /R84 80 0 R /F2 215 0 R /R119 178 0 R /R153 216 0 R /R29 40 0 R /R88 97 0 R /R37 44 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 221 0 R 222 0 R 223 0 R 224 0 R 225 0 R 226 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 227 0 R /Resources << /ExtGState << /R28 29 0 R >> /Font << /R181 228 0 R /R78 84 0 R /F1 232 0 R /R76 93 0 R /F2 233 0 R /R125 153 0 R /R88 97 0 R /R123 157 0 R /R135 161 0 R /R121 165 0 R /R90 101 0 R /R80 109 0 R /R86 112 0 R /R84 80 0 R /R119 178 0 R /R179 234 0 R /R175 237 0 R /R177 240 0 R /R171 244 0 R /R173 248 0 R /R29 40 0 R /R133 185 0 R /R37 44 0 R /R33 51 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 251 0 R 252 0 R 253 0 R 254 0 R 255 0 R 256 0 R 257 0 R 258 0 R 259 0 R 260 0 R 261 0 R 262 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 263 0 R /Resources << /ColorSpace << /R206 264 0 R >> /XObject << /R215 265 0 R /R214 266 0 R /R226 267 0 R /R232 268 0 R /R231 269 0 R /R225 270 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /Pattern << /R211 271 0 R /R210 273 0 R /R213 274 0 R /R223 275 0 R /R217 276 0 R /R203 277 0 R /R220 278 0 R /R221 279 0 R /R222 280 0 R /R209 281 0 R /R224 282 0 R /R212 283 0 R >> /ExtGState << /R202 272 0 R /R201 284 0 R /R216 285 0 R >> /Font << /R229 286 0 R /R119 178 0 R /R207 289 0 R /R204 292 0 R /R218 295 0 R /F2 298 0 R /R227 299 0 R /R29 40 0 R /R125 153 0 R /F1 302 0 R /R123 157 0 R /R121 165 0 R /R37 44 0 R /R84 80 0 R /R31 55 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 303 0 R 304 0 R 305 0 R 306 0 R 307 0 R 308 0 R 309 0 R 310 0 R 311 0 R 312 0 R 313 0 R 314 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 315 0 R /Resources << /ExtGState << /R201 284 0 R /R216 285 0 R >> /Font << /R264 316 0 R /F2 319 0 R /R266 320 0 R /R260 323 0 R /R262 326 0 R /R76 93 0 R /R29 40 0 R /F1 329 0 R /R123 157 0 R /R258 330 0 R /R121 165 0 R /R37 44 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 333 0 R 334 0 R 335 0 R 336 0 R 337 0 R 338 0 R 339 0 R 340 0 R 341 0 R 342 0 R 343 0 R 344 0 R 345 0 R 346 0 R 347 0 R 348 0 R 349 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 350 0 R /Resources << /ExtGState << /R201 284 0 R /R216 285 0 R >> /Font << /R310 351 0 R /R312 354 0 R /R76 93 0 R /F2 357 0 R /R308 358 0 R /R123 157 0 R /R121 165 0 R /R302 361 0 R /R300 364 0 R /R306 367 0 R /R304 370 0 R /R298 373 0 R /R119 178 0 R /R294 376 0 R /R296 379 0 R /R29 40 0 R /F1 382 0 R /R37 44 0 R /R84 80 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 383 0 R 384 0 R 385 0 R 386 0 R 387 0 R 388 0 R 389 0 R 390 0 R 391 0 R 392 0 R 393 0 R 394 0 R 395 0 R 396 0 R 397 0 R 398 0 R 399 0 R 400 0 R 401 0 R 402 0 R 403 0 R 404 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 405 0 R /Resources << /ExtGState << /R28 29 0 R >> /Font << /R29 40 0 R /R37 44 0 R /F1 406 0 R /F2 407 0 R /R31 55 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 16442 >> stream q q q 0.1 0 0 0.1 0 0 cm /R28 gs 0 g q 10 0 0 10 0 0 cm BT /R29 14.3462 Tf 1 0 0 1 65.141 675.067 Tm [ (Unsuper) 10 (vised) -249.989 (Lear) 14.9893 (ning) -249.993 (and) -249.991 (Segmentation) -250.002 (of) -250.012 (Complex) -249.99 (Acti) 10.0099 (vities) -249.991 (fr) 18.0082 (om) -250.008 (V) 37.0137 (ideo) ] TJ /R31 11.9552 Tf 168.655 -37.8582 Td [ (F) 15.0158 (adime) -250.01 (Sener) 40.0033 (\054) -250.012 (Angela) -249.985 (Y) 99.9847 (ao) ] TJ -7.24492 -13.9469 Td [ (Uni) 24.9957 (v) 14.9851 (ersity) -249.989 (of) -250.014 (Bonn\054) -249.987 (German) 14.9872 (y) ] TJ /R33 8.9664 Tf 1.89492 -13.948 Td [ (f) -0.90126 ] TJ /R35 8.9664 Tf 4.60781 0 Td (sener\054yao) Tj /R33 8.9664 Tf 48.418 0 Td [ (g) -0.90126 ] TJ /R35 8.9664 Tf 4.6082 0 Td (\100cs\056uni\055bonn\056de) Tj /R29 11.9552 Tf -140.085 -41.0461 Td (Abstract) Tj /R37 9.9626 Tf -83.9281 -24.525 Td [ (This) -256.003 (paper) -256.999 (pr) 36.9852 (esents) -255.981 (a) -257.011 (ne) 15.0183 (w) -255.983 (method) -256.995 (for) -256.019 (unsupervised) -257.018 (se) 39.9958 (g\055) ] TJ -11.9547 -11.9559 Td [ (mentation) -452.983 (of) -451.991 (comple) 20.0077 (x) -452.988 (activities) -452.988 (fr) 44.9864 (om) -451.996 (video) -453.015 (into) -452.998 (multiple) ] TJ 11.9551 TL T* [ (steps\054) -322.982 (or) -308.996 (sub\055activities\054) -324.015 (wi) 1.00106 (thout) -309.017 (any) -309.001 (te) 20 (xtual) -308.988 (input\056) -486.005 (W) 91.9871 (e) -309.012 (pr) 44.9851 (o\055) ] TJ T* [ (pose) -233.008 (an) -233.01 (iter) 14.9938 (ative) -233.009 (discriminative\055g) 10.0143 (ener) 15.0196 (ative) -233.01 (appr) 44.9949 (oac) 14.9828 (h) -233.015 (whic) 15 (h) ] TJ T* [ (alternates) -446.984 (between) -447.006 (discriminatively) -447.008 (learning) -448.006 (the) -447.016 (appear) 20.0187 (\055) ] TJ T* [ (ance) -426.006 (of) -425.995 (sub\055activities) -426 (fr) 44.9864 (om) -426 (the) -426 (videos\047) -427.015 (visual) -425.98 (featur) 37.0012 (es) -426.005 (to) ] TJ T* [ (sub\055activity) -373.982 (labels) -373.01 (and) -374.001 (g) 10.0032 (ener) 15.0196 (atively) -372.983 (modelling) -374.016 (the) -374.009 (tempo\055) ] TJ 11.9563 TL T* [ (r) 14.984 (al) -269.982 (structur) 37.0128 (e) -269.998 (of) -271.001 (sub\055activities) -269.987 (using) -270.005 (a) -269.989 (Gener) 14.9828 (alized) -270.987 (Mallows) ] TJ 11.9547 TL T* [ (Model\056) -518.014 (In) -319.004 (addition\054) -336.013 (we) -320.016 (intr) 44.9986 (oduce) -319 (a) -319 (model) -318.991 (for) -320.008 (bac) 20.0016 (kgr) 45.0194 (ound) ] TJ T* [ (to) -227.989 (account) -227.983 (for) -227.983 (fr) 14.9914 (ames) -229.011 (unr) 36.9902 (elat) 0.99493 (ed) -228.98 (to) -227.989 (the) -227.994 (actual) -228.016 (activities\056) -303.011 (Our) ] TJ T* [ (appr) 44.9937 (oac) 14.9834 (h) -261.01 (is) -261.002 (validated) -260.982 (on) -261.005 (the) -261.008 (c) 15.0122 (hallenging) -261.008 (Br) 36.984 (eakfast) -260.993 (Actions) ] TJ T* [ (and) -455.007 (Inria) -455.982 (Instructional) -455.003 (V) 73.9913 (ideos) -455.004 (datasets) -454.985 (and) -455.988 (outperforms) ] TJ T* [ (both) -351.998 (unsupervised) -351.982 (and) -352.985 (weakly\055supervised) -351.983 (state) -351.998 (of) -352.007 (the) -352.993 (art\056) ] TJ /R29 11.9552 Tf 24.1539 TL T* [ (1\056) -249.99 (Intr) 18.0146 (oduction) ] TJ /R31 9.9626 Tf 15.0719 -19.543 Td [ (W) 79.9873 (e) -313.011 (address) -312.997 (the) -312.019 (problem) -313.019 (of) -313.005 (understanding) -313.002 (comple) 14.9975 (x) -313.002 (ac\055) ] TJ -15.0719 -11.9559 Td [ (ti) 24.9903 (vities) -280.004 (from) -279.982 (video) -280.994 (sequences\056) -400 (A) -281.009 (comple) 14.9975 (x) -279.988 (acti) 24.9811 (vity) -280.007 (is) -280.019 (a) -281.017 (pro\055) ] TJ 11.9551 TL T* [ (cedural) -228 (task) -227.007 (with) -228.018 (multiple) -227.014 (steps) -227.982 (or) -228 (sub\055acti) 24.9958 (vities) -226.992 (that) -227.986 (follo) 24.9983 (w) ] TJ T* [ (some) -219.984 (loose) -219.984 (ordering\056) -300.01 (Comple) 15.0159 (x) -221.017 (acti) 24.9824 (vities) -220.012 (can) -220.005 (be) -220.003 (found) -219.985 (in) -221.01 (in\055) ] TJ T* [ (structional) -226.003 (videos\073) -234.989 (Y) 109.984 (ouT) 44.9864 (ube) -225.995 (hosts) -226.011 (hundreds) -226.996 (of) -226 (thousands) -226.982 (of) ] TJ T* [ (such) -306.996 (videos) -306.003 (on) -306.998 (acti) 24.9824 (vities) -307.019 (as) -305.987 (common) -307.011 (as) ] TJ /R37 9.9626 Tf 162.813 0 Td [ (\140making) -306.99 (cof) 18.0043 (fee\047) ] TJ /R31 9.9626 Tf 65.6859 0 Td (to) Tj -228.499 -11.9563 Td [ (the) -269.006 (more) -269.984 (obscure) ] TJ /R37 9.9626 Tf 71.6828 0 Td [ (\140weaving) -268.991 (banana) -270.003 <026272> 36.9914 (e) -269.019 (cloths\047) ] TJ /R31 9.9626 Tf 119.474 0 Td [ (\056) -368.015 (Similarly) 64.9818 (\054) ] TJ -191.157 -11.9547 Td [ (in) -302.016 (assisti) 25.0118 (v) 14.9828 (e) -301.993 (robotics\054) -315.015 (a) -301.013 (robot) -302.01 (that) -302.013 (can) -301.992 (understand) -302.001 (and) -301.984 (parse) ] TJ T* [ (the) -338.014 (steps) -338.982 (of) -337.981 (a) -337.988 (household) -339.002 (task) -338.009 (such) -338.99 (as) ] TJ /R37 9.9626 Tf 155.356 0 Td [ (\140doing) -337.998 (laundry\047) ] TJ /R31 9.9626 Tf 67.066 0 Td (can) Tj -222.422 -11.9551 Td [ (anticipate) -250.015 (and) -249.991 (support) -250.006 (upcoming) -249.996 (steps) -250.017 (or) -249.995 (sub\055acti) 24.9958 (vities\056) ] TJ 11.9551 -12.5699 Td [ (Comple) 15.0165 (x) -442 (acti) 24.9824 (vity) -442.019 (understanding) -442 (has) -441.997 (recei) 25.0056 (v) 14.9828 (ed) -442.005 (little) -442.012 (at\055) ] TJ -11.9551 -11.9551 Td [ (tention) -490.984 (in) -491.004 (the) -491.009 (computer) -490.994 (vision) -490.981 (community) -491.001 (compared) -491.006 (to) ] TJ T* [ (the) -368.009 (more) -368.987 (popular) -367.997 (simple) -367.984 (action) -368.981 (recognition) -367.983 (task\056) -665.009 (In) -368.995 (sim\055) ] TJ T* [ (ple) -386.006 (action) -385.998 (recognition\054) -420.991 (short\054) -419.986 (trimmed) -386.016 (clips) -385.992 (are) -387.007 <636c61737369026564> ] TJ 11.9559 TL T* [ (with) -407.988 (single) -407.004 (labels\054) ] TJ /R37 9.9626 Tf 79.8258 0 Td [ (e) 15.0122 (\056g) ] TJ /R31 9.9626 Tf 11.7461 0 Td [ (\056) -408.008 (of) -408.009 (sports\054) -446.995 (playing) -407 (musical) -407.981 (instru\055) ] TJ -91.5719 -11.9551 Td [ (ments) -343.014 (\13310\054) -342.992 (27\135\054) -366.008 (and) -342.995 (so) -342.991 (on\056) -588.987 (Performance) -342.992 (on) -342.992 (simple) -343.007 (action) ] TJ 11.9547 TL T* [ (recognition) -418.993 (has) -420 (seen) -418.986 (a) -418.994 (remarkable) -418.983 (boost) -420.018 (with) -419.006 (the) -418.982 (use) -420 (of) ] TJ T* [ (deep) -275.012 (architectures) -275.003 (\13310\054) -275.983 (25\054) -275 (29\135\056) -386.004 (Such) -274.993 (methods) -275.015 (ho) 24.986 (we) 25.013 (v) 14.9828 (er) -276 (are) ] TJ T* [ (rarely) -313.99 (applicable) -314.987 (for) -313.988 (temporally) -313.989 (localizing) -313.994 (and\057or) -314.992 (classify\055) ] TJ 11.9559 TL T* [ (ing) -318.008 (actions) -316.984 (from) -318.014 (longer) 40.008 (\054) -333.981 (untrimmed) -318.018 (video) -316.989 (sequences\054) -334.991 (usu\055) ] TJ 11.9551 TL T* [ (ally) -276.997 (due) -275.988 (to) -277 (the) -277.005 (lack) -275.989 (of) -277.011 (temporal) -276.988 (consideration\056) -390.013 (Ev) 14.9877 (en) -277.013 (w) 10 (orks) ] TJ T* [ (which) -411.994 (do) -412 (incorporate) -411.995 (some) -411.991 (modelling) -412 (of) -412.007 (temporal) -411.985 (struc\055) ] TJ T* [ (ture) -307.984 (\1334\054) -307.984 (24\054) -308.015 (28\054) -308.015 (29\135) -307 (do) -308.017 (little) -307.995 (more) -308.017 (than) -308.015 (capturing) -308.005 (frame\055to\055) ] TJ T* [ (frame) -293.993 (changes\054) -306.016 (which) -294.994 (is) -294.016 (wh) 4.98691 (y) -295.005 (the) -293.983 (state) -294.997 (of) -293.988 (the) -295.002 (art) -293.99 (still) -295.014 (relies) ] TJ 258.75 487.267 Td [ (on) -396.983 (either) -398.005 (optical) -397.01 <036f> 24.9958 (w) -398.01 (\13325\135) -396.988 (or) -396.993 (dense) -398.002 (trajectories) -397.017 (\13329\054) -398.002 (30\135\056) ] TJ T* [ (Mo) 14.995 (ving) -292.008 (to) 24.9885 (w) 10 (ards) -291.988 (understanding) -291.005 (comple) 14.9975 (x) -291.986 (acti) 24.9811 (vities) -292 (then) -292.015 (be\055) ] TJ T* [ (comes) -296.014 (e) 25.0105 (v) 14.9828 (en) -295.99 (more) -296.019 (challenging\054) -307.012 (as) -295.99 (it) -296.014 (requires) -295.98 (not) -296.009 (only) -296.005 (pars\055) ] TJ T* [ (ing) -213.006 (long) -213.981 (video) -213.006 (sequences) -213.994 (into) -212.999 (semantically) -212.984 (meaningful) -214.008 (sub\055) ] TJ T* [ (acti) 24.9811 (vities\054) -266.002 (b) 20.0016 (ut) -263.003 (also) -263 (capturing) -262.99 (the) -263.005 (temporal) -262.99 (relationships) -263.005 (that) ] TJ T* [ (occur) -250 (between) -249.99 (these) -249.988 (sub\055acti) 24.9958 (vities\056) ] TJ 11.9551 -12.7402 Td [ (W) 79.9866 (e) -287.996 (aim) -288.016 (to) -288.018 (disco) 14.9975 (v) 14.9828 (er) -287.999 (and) -287.986 (se) 15.0171 (gment) -288.006 (the) -287.981 (steps) -288.011 (of) -287.991 (a) -287.996 (comple) 14.9975 (x) ] TJ -11.9551 -11.9559 Td [ (acti) 24.9811 (vity) -343.997 (from) -344.011 (collections) -343.984 (of) -344.021 (video) -344.001 (in) -344.011 (an) -343.982 (unsupervised) -343.982 (w) 10.0032 (ay) ] TJ T* [ (based) -424.981 (purely) -425.012 (on) -423.998 (visual) -425 (inputs\056) -834.018 (W) 39.9933 (ithin) -424.998 (the) -425.017 (same) -425.017 (acti) 24.9811 (vity) ] TJ T* [ (class\054) -231.988 (it) -227.006 (is) -227.009 (lik) 10.0179 (ely) -227.994 (that) -227.006 (videos) -226.996 (share) -228.001 (common) -226.987 (steps) -227.001 (and) -227.996 (follo) 24.9958 (w) ] TJ T* [ (a) -212.009 (similar) -210.992 (temporal) -212.019 (ordering\056) -296.99 (T) 79.9916 (o) -211.999 (date\054) -219.998 (w) 10.0032 (orks) -211.009 (in) -211.994 (a) -212.009 (similar) -212.009 (v) 14.9828 (ein) ] TJ T* [ (of) -317.003 (unsupervised) -317.008 (learning) -316.009 (all) -316.996 (require) -317.006 (inputs) -317.011 (from) -316.996 (narration\073) ] TJ T* [ (the) -203.998 (sub\055acti) 24.9958 (vities) -204.015 (and) -203.02 (sequence) -204 (information) -204 (are) -204.015 (e) 15.0122 (xtracted) -203.995 (ei\055) ] TJ 11.9559 TL T* [ (ther) -247.013 (entirely) -246.998 (from) -247.008 (\13317\054) -247.008 (1\135\054) -247.993 (or) -247.018 (rely) -247.013 (hea) 19.9918 (vily) -246.993 (\13323\135) -247.993 (on) -247.008 (te) 14.9828 (xt\056) -308.993 (Such) ] TJ 11.9551 TL T* [ (w) 10 (orks) -342.009 (assume) -340.982 (that) -342.007 (the) -340.995 (te) 14.9828 (xt) -342.012 (is) -340.987 (well\055aligned) -341.992 (with) -341.017 (the) -342.012 (visual) ] TJ T* [ (information) -380.99 (of) -380.993 (the) -382.007 (video) -381.015 (so) -380.985 (that) -381.02 (visual) -381.99 (repres) 1.01454 (entations) -381.98 (of) ] TJ T* [ (the) -275.006 (sub\055acti) 24.9958 (vity) -274.988 (are) -276.005 (learned) -275.013 (from) -275.003 (within) -275.018 (the) -275.983 (t) 0.98513 (e) 15.0122 (x) -1.01454 (t) 0.98513 (\047) 55.0055 (s) -275.988 (temporal) ] TJ T* [ (bounds\056) -303.013 (This) -226.999 (is) -227.989 (not) -227.984 (al) 10 (w) 10 (ays) -227.996 (the) -227.011 (case) -228.006 (for) -228.001 (instructional) -227.991 (videos\054) ] TJ 11.9559 TL T* [ (as) -237.017 (it) -235.985 (is) -237.007 (f) 9.99588 (ar) -236.987 (more) -235.99 (natural) -237.007 (for) -236 (the) -237.009 (human) -236.995 (narrator) -236 (to) -237.009 <02727374> -236.995 (speak) ] TJ 11.9551 TL T* [ (about) -284.994 (what) -285.001 (will) -284.987 (be) -285.011 (done\054) -292.998 (and) -285.006 (then) -284.997 (carry) -284.982 (out) -284.992 (the) -285.001 (action\056) -415.009 (Fi\055) ] TJ T* [ (nally) 64.9892 (\054) -249.995 (reliably) -250.997 (parsing) -250.017 (spok) 10.0179 (en) -251.017 (natural) -249.983 (language) -249.983 (into) -250.992 (scripts) ] TJ /R31 6.9738 Tf 232.265 3.61484 Td (1) Tj /R31 9.9626 Tf -232.265 -15.5699 Td [ (is) -265.002 (an) -265.015 (unsolv) 15.0122 (ed) -265.015 (and) -265.01 (open) -265.005 (research) -264.99 (topic) -266.014 (i) 0.98513 (n) -265.99 (itself\056) -355.014 (As) -265.005 (such\054) -269 (it) ] TJ T* [ (is) -249.985 (in) -249.985 (our) -249.99 (interest) -250.017 (to) -249.985 (rely) -249.993 (only) -250.012 (on) -249.988 (visual) -250.012 (inputs\056) ] TJ 11.9551 -12.7398 Td [ (In) -376.993 (this) -377.991 (w) 10 (ork\054) -408.981 (we) -378.018 (propose) -377.016 (an) -376.996 (iterati) 24.9909 (v) 14.9828 (e) -377.981 (model) -377.011 (which) -378 (al\055) ] TJ -11.9551 -11.9551 Td [ (ternates) -227.991 (between) -229.013 (learning) -227.982 (a) -228.986 (discriminati) 24.9909 (v) 14.9828 (e) -228.006 (representation) -228.996 (of) ] TJ 11.9559 TL T* [ (a) -341.007 (video\047) 54.9859 (s) -340.997 (visual) -341.014 (features) -341.004 (to) -340.992 (sub\055acti) 24.9958 (vities) -341.012 (and) -340.997 (a) -341.007 (generati) 24.986 (v) 14.9828 (e) ] TJ 11.9551 TL T* [ (model) -381.01 (of) -382.012 (the) -380.988 (sub\055acti) 24.9958 (vities\047) -381.01 (temporal) -381.01 (structure\056) -703.981 (By) -381.995 (com\055) ] TJ T* [ (bining) -438.01 (the) -439.017 (sub\055acti) 24.9958 (vity) -437.981 (representations) -438.015 (with) -437.981 (the) -439.015 (temporal) ] TJ T* [ (model\054) -327.997 (we) -313.009 (arri) 25.013 (v) 14.9828 (e) -313.012 (at) -311.985 (a) -313.012 (se) 15.0196 (gmentation) -313.002 (of) -311.987 (the) -312.997 (video) -312.987 (sequence\054) ] TJ T* [ (which) -259 (is) -257.984 (then) -259.003 (used) -259.003 (to) -259.003 (update) -259 (the) -257.986 (visual) -258.991 (representations) -258.986 (\050see) ] TJ T* [ (Fig\056) -202.996 (1a\051\056) -293.99 (W) 79.9866 (e) -202.991 (represent) -202 (sub\055acti) 24.9958 (vities) -202.996 (by) -201.996 (learning) -203.005 (linear) -203.015 (map\055) ] TJ 11.9563 TL T* [ (pings) -252.007 (from) -253.007 (visual) -252.009 (features) -251.997 (to) -253.007 (a) -252.002 (lo) 24.986 (w) -251.992 (dimensional) -253.007 (embedding) ] TJ 11.9547 TL T* [ (space) -202.008 (with) -201.983 (a) -202.991 (ranking) -201.986 (loss\056) -294.005 (The) -202 (mappings) -202.006 (are) -202.996 (opti) 1.00473 (mized) -203.01 (such) ] TJ T* [ (that) -336.008 (visual) -335.995 (features) -335.986 (from) -336.013 (the) -336.013 (same) -336.013 (sub\055acti) 24.9958 (vity) -335.998 (are) -335.993 (pushed) ] TJ T* [ (together) 40.0105 (\054) -249.995 (while) -250.02 (dif) 24.986 (ferent) -250 (sub\055acti) 24.9958 (vities) -250.007 (are) -250.007 (pulled) -250.012 (apart\056) ] TJ 11.9551 -12.7398 Td [ (T) 69.9932 (emporally) 64.999 (\054) -400.989 (we) -372.019 (treat) -370.994 (a) -371.002 (comple) 14.9975 (x) -370.992 (acti) 24.9811 (vity) -371.012 (as) -372.016 (a) -371.002 (sequence) ] TJ -11.9551 -11.9551 Td [ (of) -553.003 (permutable) -552.001 (sub\055acti) 24.9958 (vities) -553.016 (and) -551.981 (model) -552.981 (the) -552.996 (distrib) 19.9918 (ution) ] TJ 11.9563 TL T* [ (o) 14.9828 (v) 14.9828 (er) -601.005 (permutations) -600.988 (with) -601.015 (a) -601.003 (Generalized) -600.998 (Mallo) 25 (ws) -600.988 (Model) ] TJ 11.9547 TL T* [ (\050GMM\051) -199.986 (\1335\135\056) -293.01 (GMMs) -200.013 (ha) 19.9967 (v) 14.9828 (e) -200.011 (been) -200.006 (s) 0.98513 (uccessfully) -199.981 (used) -199.996 (in) -199.996 (the) -199.996 (NLP) ] TJ ET Q 3.98 w 0 G 3088.62 1001.43 m 4033.61 1001.43 l S q 10 0 0 10 0 0 cm BT /R31 5.9776 Tf 1 0 0 1 319.721 93.277 Tm (1) Tj /R31 7.9701 Tf 3.48711 -2.81211 Td [ (Here\054) -375.978 (we) -350.998 (refer) -351.004 (to) -351.025 (t) 0.98942 (he) -350.976 (NLP) -351.013 <6465026e6974696f6e> -350.976 (of) -351.013 (script) -350.014 (as) ] TJ /R37 7.9701 Tf 162.513 0 Td [ (\223a) -351.019 (pr) 36.9852 (edetermined\054) ] TJ -176.859 -9.46484 Td [ (ster) 37.0127 (eotyped) -250.019 (sequence) -250.022 (of) -250.013 (actions) -249.976 (that) -250.019 <6465026e65> -249.982 (a) -250.019 (well\055known) -249.988 (situation\224) ] TJ /R31 7.9701 Tf 217.319 0 Td (\13322\135\056) Tj ET Q Q Q q q 1 1 1 rg /a0 gs 48.406 786.422 515.188 -52.699 re f q /s5 gs /x6 Do Q q /s7 gs /x8 Do Q q /s9 gs /x10 Do Q q /s11 gs /x12 Do Q Q Q Q q 1 0 0 1 0 0 cm BT /F1 12 Tf 14.4 TL ET 1 1 1 rg n 270 32 72 14 re f* 0.5 0.5 0.5 rg BT /F2 9 Tf 10.8 TL ET BT 1 0 0 1 297 35 Tm (8368) Tj T* ET Q endstream endobj 14 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 15 0 R >> >> /Length 28 /Group << /Type /Group /S /Transparency /CS /DeviceRGB /I true >> /BBox [ 78 746 96 765 ] /Type /XObject /Subtype /Form >> stream x+O4PH/VЯ0Pp 0 endstream endobj 15 0 obj << /Filter /FlateDecode /Resources 16 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
16 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
17 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 18 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 67 752 84 775 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ04Up
0
endstream
endobj
18 0 obj
<<
/Filter /FlateDecode
/Resources 19 0 R
/Length 228
/Type /XObject
/BBox [ 67 752 84 775 ]
/Subtype /Form
>>
stream
xeQKn!s ?FPav6R٪TS.
b];15YyR
{7QL.\:Rv/x9l+L7h%1!}i/AI(kz"U&,YO![R hg{3}4/GyYF:!w}Gn+'xJcO9i뽼_-:`
endstream
endobj
19 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
20 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 21 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 132 751 480 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources 22 0 R
/Length 53223
/Type /XObject
/BBox [ 132 751 480 772 ]
/Subtype /Form
>>
stream
xtI:6%Q㨈?7rA= u%6 ?Y(WbWo{B>9
x`Znϳ|8{3?0x*z ǃ|,@:w>`c|*ϻⳅKO3`g
:_|}}><.6`Z{{3]#<_o"~:ͺgk7/Ұ@|K yp ]03ʷCmş8˽Y?>(3!Bwqs.Z8,~~=rMT̩y+/*w: uBZ_`ߵp`%M?ɝ1ɳw=vDۉy&xb4Q>d@ sg~lA