%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 (MSR\055VTT\072 A Large Video Description Dataset for Bridging Video and Language) /Producer (PyPDF2) /Author (Jun Xu\054 Tao Mei\054 Ting Yao\054 Yong Rui) /Subject (2016 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /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 /s11 29 0 R /a0 << /CA 1 /ca 1 >> /R7 32 0 R /s5 33 0 R /s7 36 0 R >> /Font << /F2 39 0 R /R12 40 0 R /F1 45 0 R /R14 46 0 R /R16 51 0 R /R18 55 0 R /R20 60 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] /ColorSpace << /R11 65 0 R /R9 66 0 R >> >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 69 0 R /Resources << /XObject << /R49 70 0 R /R48 71 0 R /R47 72 0 R /R55 73 0 R /R45 74 0 R /R44 75 0 R /R43 76 0 R /R42 77 0 R /R41 78 0 R /R40 79 0 R /R52 80 0 R /R53 81 0 R /R50 82 0 R /R39 83 0 R /R54 84 0 R /R29 85 0 R /R28 86 0 R /R38 87 0 R /R46 88 0 R /R36 89 0 R /R37 90 0 R /R51 91 0 R /R30 92 0 R /R31 93 0 R >> /ExtGState << /R7 32 0 R >> /Font << /F2 94 0 R /R12 40 0 R /F1 95 0 R /R14 46 0 R /R33 96 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R11 65 0 R /R9 66 0 R >> >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 103 0 R /Resources << /XObject << /R63 104 0 R >> /ExtGState << /R64 105 0 R /R7 32 0 R >> /Font << /F1 106 0 R /F2 107 0 R /R12 40 0 R /R66 108 0 R /R14 46 0 R /R20 60 0 R /R33 96 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R11 65 0 R /R9 66 0 R >> >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 115 0 R /Resources << /ColorSpace << /R11 65 0 R /R9 66 0 R >> /ExtGState << /R76 116 0 R >> /Font << /F2 117 0 R /R12 40 0 R /F1 118 0 R /R14 46 0 R /R77 119 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 122 0 R /Resources << /XObject << /R101 123 0 R /R100 124 0 R /R102 125 0 R /R99 126 0 R /R91 127 0 R /R92 128 0 R /R93 129 0 R /R94 130 0 R /R86 131 0 R /R85 132 0 R >> /ExtGState << /R64 105 0 R /R7 32 0 R >> /Font << /R103 133 0 R /R105 138 0 R /R77 119 0 R /R107 141 0 R /F1 145 0 R /F2 146 0 R /R12 40 0 R /R66 108 0 R /R14 46 0 R /R88 147 0 R /R96 154 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R11 65 0 R /R9 66 0 R >> >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 161 0 R /Resources << /ColorSpace << /R11 65 0 R /R9 66 0 R >> /ExtGState << /R76 116 0 R >> /Font << /R103 133 0 R /R116 162 0 R /R77 119 0 R /F2 166 0 R /R12 40 0 R /F1 167 0 R /R14 46 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 168 0 R /Resources << /ColorSpace << /R127 169 0 R /R11 65 0 R /R125 170 0 R /R9 66 0 R >> /Pattern << /R126 171 0 R /R129 173 0 R /R128 174 0 R >> /ExtGState << /R64 105 0 R /R124 172 0 R /R7 32 0 R >> /Font << /F2 175 0 R /R12 40 0 R /R66 108 0 R /R14 46 0 R /F1 176 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 177 0 R /Resources << /XObject << /R139 178 0 R /R163 179 0 R /R162 180 0 R /R161 181 0 R /R160 182 0 R /R141 183 0 R /R166 184 0 R /R165 185 0 R /R164 186 0 R /R168 187 0 R /R149 188 0 R /R148 189 0 R /R145 190 0 R /R144 191 0 R /R147 192 0 R /R146 193 0 R /R137 194 0 R /R138 195 0 R /R167 196 0 R /R140 197 0 R /R156 198 0 R /R157 199 0 R /R154 200 0 R /R155 201 0 R /R152 202 0 R /R153 203 0 R /R150 204 0 R /R151 205 0 R /R142 206 0 R /R158 207 0 R /R159 208 0 R /R143 209 0 R >> /ExtGState << /R64 105 0 R /R7 32 0 R >> /Font << /F2 210 0 R /R12 40 0 R /R66 108 0 R /R14 46 0 R /R103 133 0 R /F1 211 0 R /R169 212 0 R /R33 96 0 R /R77 119 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R11 65 0 R /R9 66 0 R >> >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /TrimBox [ 0 0 612 792 ] /Contents 215 0 R /Resources << /ColorSpace << /R11 65 0 R /R9 66 0 R >> /ExtGState << /R7 32 0 R >> /Font << /F2 216 0 R /R20 60 0 R /R12 40 0 R /F1 217 0 R /R14 46 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /CropBox [ 0 0 612 792 ] /MediaBox [ 0 0 612 792 ] /Annots [ ] /Type /Page /BleedBox [ 0 0 612 792 ] /ArtBox [ 0 0 612 792 ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 22879 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R7 gs /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R12 14.34620 Tf 1 0 0 1 50.11210 675.06700 Tm [ (MSR\055VTT) 74.00270 (\072) -241.00300 (A) -240.98900 (Lar) 9.99370 (ge) -241.01000 (V) 37.01370 (ideo) -240.99100 (Description) -241.00800 (Dataset) -240.99100 (f) 24.99230 (or) -240.98700 (Bridging) -241.00900 (V) 37.01370 (ideo) -240.98900 (and) -241.00600 (Language) ] TJ /R14 11.95520 Tf 145.17600 -37.85820 Td [ (Jun) -250.09500 (Xu) -250 (\054) -250.01200 (T) 80.01470 (ao) -250.00800 (Mei) -249.98300 (\054) -250.01200 (T) 34.98980 (ing) -250.00400 (Y) 99.98470 (ao) -250.00200 (and) -249.98700 (Y) 110.01600 (ong) -249.98900 (Rui) ] TJ 17.48480 -13.94690 Td [ (Microsoft) -250.10800 (Research\054) -250.09300 (Beijing\054) -249.98900 (China) ] TJ /R16 8.96640 Tf -35.43480 -13.94800 Td (\173) Tj /R18 8.96640 Tf 4.60781 0 Td [ (v\055junfu\054) -599.89000 (tmei\054) -599.89900 (tiyao\054) -600.01000 (yongrui) ] TJ /R16 8.96640 Tf 156.01500 0 Td (\175) Tj /R18 8.96640 Tf 4.60703 0 Td (\100microsoft\056com) Tj /R12 11.95520 Tf -196.57300 -41.04610 Td (Abstract) Tj /R20 9.96260 Tf -83.92810 -24.25820 Td [ (While) -372 (ther) 36.98770 (e) -371.98200 (has) -372 (been) -372.01500 (incr) 36.98770 (easing) -371.99200 (inter) 36.99630 (est) -372.00700 (in) -372.98400 (the) -372.00900 (task) -372.00200 (of) ] TJ -11.95510 -11.95590 Td [ (describing) -337.00400 (video) -338.01300 (with) -337.00900 (natur) 15.00610 (al) -338.04900 (langu) ] TJ -0.04981 Tc 146.14200 0 Td [ (age\054) -364.01300 (curr) ] TJ 0 Tc 36.94140 0 Td [ (ent) -338.01500 (computer) ] TJ -183.08400 -11.95510 Td [ (vision) -355.01200 (algorithms) -353.88500 (ar) 36.98520 (e) -355.00400 (still) -355.00400 (se) 15.01830 (ver) 37.99120 (ely) -355 (limited) -355.00700 (in) -354.00700 (terms) -354.98200 (of) -354.98700 (the) ] TJ 11.95510 TL T* [ (variability) -271.90600 (and) -272.99900 (comple) 20.08610 (xity) -271.98800 (of) -273.00100 (the) -271.98600 (videos) -272.99900 (and) -272.01800 (their) -272.99100 (associat\055) ] TJ T* [ (ed) -287.99100 (langua) 9.99159 (g) 10.00320 (e) -287.01500 (that) -288.00600 (the) 30.00230 (y) -287.01600 (can) -287.98500 (r) 37.01960 (eco) 9.99466 (gnize) 14.99630 (\056) -422.98600 (This) -287.01800 (is) -288.01800 (in) -288.01800 (part) -288.00800 (due) -287.98600 (to) ] TJ T* [ (the) -219.01400 (simplicity) -218.98100 (of) -219.01500 (curr) 36.98650 (ent) -217.99500 (benc) 15.01830 (hmarks\054) -224.99800 (whic) 15 (h) -219.01700 (mostly) -218.01800 (focus) -219.00300 (on) ] TJ 11.95590 TL T* [ (speci\223c) -310.99700 (\223ne\055gr) 14.99140 (ained) -309.89100 (domains) -311.01100 (with) -311.01300 (limited) -309.99500 (videos) -310.99200 (and) -310.99200 (sim\055) ] TJ T* [ (ple) -324.01600 (descriptions\056) -531.89700 (While) -324.00700 (r) 37.01960 (esear) 36.98280 (c) 15.01220 (her) 10.00570 (s) -324.01800 (have) -323.98300 (pr) 44.98510 (o) 10.00320 (vided) -324.01700 (se) 15.01960 (ver) 15.01470 (al) ] TJ T* [ (benc) 15.01830 (hmark) -217.98800 (datasets) -218.00500 (for) -217.98400 (ima) 10.01300 (g) 10.00320 (e) -218.00800 (captioning) 10.00570 (\054) -225.01900 (we) -218.00800 (ar) 37.01830 (e) -218.00800 (not) -217.98600 (awar) 37.00360 (e) ] TJ 11.95510 TL T* [ (of) -352.00700 (any) -352.99400 (lar) 36.99260 (g) 10.00320 (e\055scale) -352.01700 (video) -352.99100 (description) -352.01400 (dataset) -351.98800 (with) -353.00700 (compr) 36.98890 (e\055) ] TJ T* [ (hensive) -249.98700 (cate) 39.99700 (gories) -250.01200 (yet) -249.99900 (diver) 10.00810 (se) -249.99600 (video) -249.98900 (content\056) ] TJ 11.95510 -12.30270 Td [ (In) -263.01300 (this) -264.00900 (paper) -262.99800 (we) -263.00600 (pr) 36.98520 (esent) -263.98500 (MSR\055VTT) -263.01400 (\050standing) -263.01200 (for) -264.01700 (\215MSR\055) ] TJ -11.95510 -11.95510 Td [ (V) 73.99130 (ideo) -413.01600 (to) -413.01700 (T) 91.99870 (e) 19.99240 (xt\216\051) -413.01600 (whic) 14.99870 (h) -412.98500 (is) -412.98400 (a) -412.98500 (ne) 15.01220 (w) -413.01600 (lar) 36.99260 (g) 10.00320 (e\055scale) -412.98700 (video) -412.98300 (benc) 15.01960 (h\055) ] TJ 11.95630 TL T* [ (mark) -249.00700 (for) -250.01800 (video) -249.00900 (under) 10.01550 (standing) 10.08290 (\054) -249.99500 (especially) -248.99800 (the) -249.01000 (emer) 37.01340 (ging) -250.01500 (task) ] TJ 11.95510 TL T* [ (of) -326.01200 (tr) 14.99140 (anslating) -326.01100 (video) -326.01500 (to) -324.99200 (te) 20 (xt\056) -537.00400 (This) -325.01100 (is) -326.01100 (ac) 15.01710 (hie) 14.98520 (ved) -325.00700 (by) -325.98500 (collecting) ] TJ T* [ (257) -376.98100 (popular) -376.91800 (queries) -377.99500 (fr) 44.98640 (om) -376.98900 (a) -376.99100 (commer) 36.98280 (cial) -376.98100 (video) -376.98900 (sear) 36.98650 (c) 15.01220 (h) -378.01000 (en\055) ] TJ T* [ (gine) 10.01120 (\054) -416.00700 (with) -383.00200 (118) -383.01900 (videos) -382 (for) -383.01600 (eac) 15.01340 (h) -382.99000 (query) 55.01780 (\056) -707.99700 (In) -382.99000 (its) -383.01500 (curr) 36.98650 (ent) -382.98800 (ver) 19.99420 (\055) ] TJ T* [ (sion\054) -492.99000 (MSR\055VTT) -443.88500 (pr) 44.98390 (o) 10.00320 (vides) -443.99500 (10K) -443.98100 (web) -443.99100 (video) -443.99700 (clips) -443.98200 (with) -444.99200 (41\0562) ] TJ T* [ (hour) 10.01920 (s) -356.99300 (and) -356.98400 (200K) -357.01000 (clip\055sentence) -356.98200 (pair) 9.98240 (s) -358.01300 (in) -356.99400 (total\054) -384.01900 (co) 9.99833 (vering) -356.98500 (the) ] TJ T* [ (most) -296.00800 (compr) 36.98890 (ehensive) -295.90500 (cate) 39.91860 (gories) -295.88700 (and) -296.01400 (diver) 10.00810 (se) -297.00700 (visual) -296.00200 (content\054) ] TJ T* [ (and) -395.99700 (r) 37.01960 (epr) 36.98100 (esenting) -395.98500 (the) -396.98500 (lar) 36.99260 (g) 10.00320 (est) -396.00400 (dataset) -395.98200 (in) -396 (terms) -395.99500 (of) -397.02000 (sentence) ] TJ T* [ (and) -286.99600 (vocab) 20.00280 (ulary) 54.98960 (\056) -422.00600 (Eac) 14.98280 (h) -287.98600 (clip) -286.99600 (is) -286.99700 (annotated) -287.99100 (with) -287.01800 (about) -286.98400 (20) -287.98100 (nat\055) ] TJ T* [ (ur) 14.98890 (al) -378.00300 (sentences) -376.98400 (by) -378.01400 (1\054327) -377.99300 (AMT) -377.01600 (work) 10.01790 (er) 10 (s\056) -693.99200 (W) 91.98710 (e) -377 (pr) 36.98650 (esent) -378.00600 (a) -378.01000 (de\055) ] TJ T* [ (tailed) -373.98900 (analysis) -373.90100 (of) -374.00400 (MSR\055VTT) -374.01500 (in) -374.00400 (comparison) -374.01100 (to) -374.00400 (a) -374.01100 (complete) ] TJ 11.95590 TL T* [ (set) -364.98900 (of) -364.98600 (e) 19.99240 (xisting) -365.00400 (datasets\054) -394.01300 (to) 10.01060 (g) 10.00320 (ether) -364.00800 (with) -365.00400 (a) -364.99300 (summarization) -365.00500 (of) ] TJ 11.95510 TL T* [ (dif) 18.01660 (fer) 36.98340 (ent) -289.00300 (state\055of\055the\055art) -288.88200 (video\055to\055te) 20.01380 (xt) -289.00800 (appr) 44.99490 (oac) 14.98280 (hes\056) -427.01700 (W) 91.98710 (e) -289.01600 (also) ] TJ T* [ (pr) 44.98390 (o) 10.00320 (vide) -247.01400 (an) -247.00800 (e) 19.99240 (xtensive) -247.98900 (e) 15.01220 (valuation) -247.01500 (of) -247.00500 (these) -247.00800 (appr) 44.99490 (oac) 14.98280 (hes) -247.01000 (on) -247.98800 (this) ] TJ T* [ (dataset\054) -208.99000 (showing) -199.99800 (that) -198.88300 (the) -199.01800 (hybrid) -199.00100 (Recurr) 36.98650 (ent) -199.01800 (Neur) 15.11510 (al) -199.99400 (Network\055) ] TJ T* [ (based) -312.01000 (appr) 44.99370 (oac) 14.98400 (h\054) -328.01600 (whic) 14.99870 (h) -311.98200 (combines) -312.00400 (single\055fr) 15.01470 (ame) -311.98500 (and) -312.99200 (motion) ] TJ T* [ (r) 37.01960 (epr) 36.98100 (esentations) -346.99500 (with) -345.98800 (soft\055attention) -347.01300 (pooling) -347 (str) 14.99750 (ate) 40 (gy) 55.01040 (\054) -370.99400 (yields) ] TJ T* [ (the) -249.99000 (best) -249.98300 (g) 10.00320 (ener) 15.01960 (alization) -249.98200 (capability) -249.99600 (on) -249.98800 (MSR\055VTT) 73.98770 (\056) ] TJ /R12 11.95520 Tf 36.90660 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R14 9.96260 Tf 11.95510 -19.27700 Td [ (It) -406.98700 (has) -408.00400 (been) -406.99000 (a) -408.01500 (fundamental) -407 (yet) -408.00300 (emer) 17.99700 (ging) -407.00800 (challenge) -407.99800 (for) ] TJ -11.95510 -11.95510 Td [ (computer) -433.90500 (vision) -434.01000 (to) -433.99400 (automatically) -434 (describe) -435.01800 (visual) -434.01800 (content) ] TJ 11.95510 TL T* [ (with) -304.00600 (natural) -304.99600 (language\056) -473.00200 (Especially) 65.11540 (\054) -317.00300 (thanks) -304.98300 (to) -304.01500 (the) -303.98100 (recent) -305.01300 (de\055) ] TJ T* [ (v) 14.98280 (elopment) -444.99100 (of) -444.00300 (Recurrent) -445.01300 (Neural) -444.98100 (Netw) 10.00570 (orks) -443.99200 (\050RNNs\051\054) -494.01300 (there) ] TJ T* [ (has) -301.98200 (been) -301.98800 (tremendous) -301.91700 (interest) -302.00700 (in) -302.01600 (the) -301.98100 (task) -302.01300 (of) -301.98600 (image) -302.01300 (caption\055) ] TJ 258.75000 487.26700 Td [ (ing\054) -429.98700 (where) -392.99800 (each) -393.90200 (image) -393.99800 (is) -393.99800 (described) -392.98300 (with) -393.98800 (a) -394.01800 (single) -393.98800 (natural) ] TJ T* [ (sentence) -459.99700 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 351.06300 556.31000 Tm (7) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 356.04500 556.31000 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 363.10800 556.31000 Tm (8) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 368.09000 556.31000 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 375.16400 556.31000 Tm (11) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 385.12600 556.31000 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 392.20000 556.31000 Tm (14) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 402.16300 556.31000 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 409.23600 556.31000 Tm (22) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 419.19800 556.31000 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 426.26200 556.31000 Tm (36) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 436.22500 556.31000 Tm [ (\135\056) -940 (Along) -458.99700 (with) -460.01600 (this) -460.01600 (trend\054) ] TJ -127.36200 -11.95550 Td [ (researchers) -440.88200 (ha) 19.99670 (v) 14.98280 (e) -440.99000 (pro) 14.98520 (vided) -439.98500 (se) 25.01540 (v) 14.98280 (eral) -440.99000 (benchmark) -441.01000 (datasets) -441.00500 (to) ] TJ 11.95510 TL T* [ (boost) -403.00100 (research) -401.90600 (on) -402.98200 (image) -403.01600 (captioning) -402.01100 (\050e\056g\056\054) -441.02000 (Microsoft) -403.00100 (CO\055) ] TJ T* [ (CO) -344.98400 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 329.44500 520.44500 Tm (21) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT -0.00100 Tc /R14 9.96260 Tf 1 0 0 1 339.41700 520.44500 Tm [ (\135) -344.08500 (and) -345.09900 (Flickr) -345.10400 (30K) -345.09200 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 415.12300 520.44500 Tm (41) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 425.08400 520.44500 Tm [ (\135\051\054) -367.98300 (where) -343.98700 (tens) -344.98700 (or) -344.02100 (hundreds) -345.01600 (of) ] TJ -116.22300 -11.95510 Td [ (thousands) -249.91900 (of) -249.99500 (images) -250.01500 (are) -250.01000 (annotated) -250.01700 (with) -250.01200 (natural) -249.98300 (sentences\056) ] TJ 11.95510 -14.97580 Td [ (While) -372.00200 (there) -372.91800 (has) -372.00900 (been) -372.99400 (increasing) -371.91800 (interest) -371.99700 (in) -372.98700 (the) -372.00700 (task) -372.98200 (of) ] TJ -11.95510 -11.95510 Td [ (video) -223.00400 (to) -223.00900 (language\054) -229.00300 (e) 15.01220 (xisting) -221.99700 (approaches) -223.01700 (only) -222.99700 (achie) 25.01540 (v) 14.98280 (e) -222.98700 (se) 25.01540 (v) 14.98280 (ere\055) ] TJ 11.95510 TL T* [ (ly) -225.98900 (limited) -226 (success) -225.99200 (in) -225.00900 (terms) -225.98200 (of) -226.00100 (the) -225.99200 (v) 24.98110 (ariability) -226.00600 (and) -225.99700 (comple) 14.99750 (xity) ] TJ T* [ (of) -390.01100 (video) -389.99600 (contents) -391.00600 (and) -390.00800 (their) -390.00400 (associated) -389.99400 (language) -389.99900 (that) -391.01800 (the) 14.98770 (y) ] TJ 11.95630 TL T* [ (can) -283.01400 (recognize) -281.91100 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 370.36100 445.69200 Tm (2) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 375.34200 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 380.65200 445.69200 Tm (7) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 385.63400 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 390.94500 445.69200 Tm (15) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 400.90600 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 406.21600 445.69200 Tm (28) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 416.17900 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 421.47900 445.69200 Tm (35) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 431.44100 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 436.75200 445.69200 Tm (39) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 446.71300 445.69200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 452.02300 445.69200 Tm (34) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 461.98600 445.69200 Tm [ (\135\056) -408.01000 (This) -282.98700 (is) -282.01700 (in) -283.00200 (part) -283.00700 (due) ] TJ -153.12500 -11.95510 Td [ (to) -326.99200 (the) -326.01700 (simplicity) -326.98500 (of) -327.00200 (current) -325.98500 (benchmarks\054) -345.99600 (which) -326.99000 (mostly) -327 (fo\055) ] TJ 11.95510 TL T* [ (cus) -227.99400 (on) -229.01100 (speci\223c) -227.98200 (\223ne\055grained) -229.00100 (domains) -228.00600 (with) -228.99600 (limited) -228.00100 (data) -229.01600 (scale) ] TJ T* [ (and) -358.99400 (simple) -358.88600 (descriptions) -358.91800 (\050e\056g\056\054) -386.00900 (cooking) -359.00900 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 471.72900 409.82700 Tm (5) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 476.71100 409.82700 Tm [ (\135\054) -386.01400 (Y) 109.98700 (ouT) 44.98760 (ube) -358.99400 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 529.34300 409.82700 Tm (16) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 539.30500 409.82700 Tm (\135\054) Tj -230.44500 -11.95510 Td [ (and) -424.98300 (mo) 14.99500 (vie) -425.02000 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 359.78900 397.87200 Tm (27) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 369.75100 397.87200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 376.47600 397.87200 Tm (32) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 386.43800 397.87200 Tm [ (\135\051\056) -834.88500 (There) -424.99300 (are) -424.88000 (currently) -424.89500 (no) -425.01700 (lar) 17.99700 (ge\055scale) ] TJ -77.57890 -11.95510 Td [ (video) -294.01200 (description) -295.01200 (benchmarks) -293.89200 (that) -294.01500 (match) -294.91600 (the) -294.09800 (scale) -293.99000 (and) -295.00500 (v) 24.98110 (a\055) ] TJ 11.95630 TL T* [ (riety) -306.99500 (of) -307.00500 (e) 15.01220 (xisting) -307.00500 (image) -306.01300 (datasets) -306.98800 (because) -307.00800 (videos) -306.98300 (are) -307.01700 (signi\223\055) ] TJ 11.95510 TL T* [ (cantly) -269 (more) -269.00400 (dif) 25.10370 (\223cult) -270.00400 (and) -269.00900 (e) 15.01220 (xpensi) 25 (v) 14.98280 (e) -269.01900 (to) -269.00400 (collect\054) -274.01800 (annotate) -270.01900 (and) ] TJ -0.00100 Tc T* [ (or) 17.91610 (ganize\056) -661.09300 (Furthermore\054) -396.08100 (compared) -367.08100 (with) -367.10000 (image) -367.09100 (captioning\054) ] TJ 0 Tc T* [ (the) -245.99100 (automatic) -246.00600 (generation) -246.02000 (of) -245.01800 (video) -246.01800 (descriptions) -245.01300 (carries) -246.00300 (addi\055) ] TJ T* [ (tional) -317.99800 (challenges\054) -334.90500 (such) -318.01300 (as) -317.98600 (modeling) -319 (the) -318.01600 (spatiotemporal) -318.01600 (in\055) ] TJ T* [ (formation) -249.88700 (in) -249.98500 (video) -250.02000 (data) -249.99500 (and) -249.99300 (pooling) -249.99700 (strate) 14.98770 (gies\056) ] TJ 11.95510 -14.97580 Td [ (Moti) 25.00560 (v) 24.98110 (ated) -287.98900 (by) -289 (the) -287.98400 (abo) 14.98280 (v) 14.98280 (e) -287.99600 (observ) 24.99090 (ations\054) -298 (we) -287.99100 (present) -288.02100 (in) -289 (this) ] TJ -11.95510 -11.95510 Td [ (paper) -373 (the) -372.98900 (MSR\055VTT) -371.89100 (dataset) -372.90300 (\050standing) -373.00100 (for) -372.99700 (MSR\055V) 60.10760 (ideo) -372.98200 (to) ] TJ T* [ (T) 69.99320 (e) 15.01220 (xt\051\054) -211.01100 (which) -201.01100 (is) -201.01300 (a) -202.01000 (ne) 25.01050 (w) -200.98400 (lar) 17.99700 (ge\055scale) -200.98600 (video) -201.00600 (benchmark) -201.01100 (for) -202.00600 (video) ] TJ T* [ (understanding\054) -380.00800 (especially) -354.90200 (the) -354.01000 (emer) 17.99700 (ging) -353.99500 (task) -354.00500 (of) -355 (translating) ] TJ T* [ (video) -359.88400 (to) -360.00600 (te) 14.98030 (xt\056) -639.99600 (This) -359.99600 (is) -360.00600 (achie) 25.01540 (v) 14.98280 (ed) -360.99400 (by) -360.01300 (collecting) -359.98900 (257) -360.00400 (popular) ] TJ T* [ (queries) -339.89200 (from) -340.99200 (a) -341.00700 (commercial) -341.01700 (video) -340.98200 (search) -341.00700 (engine\054) -362.98400 (with) -341.01700 (118) ] TJ T* [ (videos) -387.00900 (for) -386.99700 (each) -387.00200 (query) 65.09450 (\056) -721.01400 (In) -386.98900 (its) -387.01400 (current) -386.99400 (v) 14.98280 (ersion\054) -421.98800 (MSR\055VTT) ] TJ T* [ (pro) 14.98520 (vides) -373.99600 (10K) -374.90600 (web) -374.01400 (video) -375.01600 (clips) -373.91800 (with) -373.99200 (41\0562) -374.89800 (hours) -373.99600 (and) -374.99100 (200K) ] TJ T* [ (clip\055sentence) -358.90300 (pairs) -358.00900 (in) -358.98700 (total\054) -385.99000 (co) 15.01710 (v) 14.98280 (ering) -358.00900 (a) -359.00400 (comprehensi) 25.00560 (v) 14.98280 (e) -359.00400 (list) ] TJ T* [ (of) -287.00800 (20) -287.98100 (cate) 15.01220 (gories) -286.99600 (and) -287.98600 (a) -287.01600 (wide) -287.98100 (v) 24.98110 (ariety) -287.00100 (of) -287.99100 (video) -286.99100 (content\056) -423.00300 (Each) ] TJ T* [ (clip) -249.98300 (w) 10 (as) -249.99300 (annotated) -250.01500 (with) -250.01500 (about) -250.01700 (20) -249.98800 (natural) -249.98300 (sentences\056) ] TJ 11.95510 -14.97500 Td [ (From) -516.98200 (a) -517.01700 (practical) -517.00700 (standpoint\054) -584.01000 (compared) -517.98200 (with) -516.98700 (e) 15.01220 (xisting) ] TJ -11.95510 -11.95510 Td [ (datasets) -247.90000 (for) -248 (video) -248.02000 (to) -247.98600 (te) 14.98030 (xt\054) -247.98300 (such) -247.98800 (as) -247.99800 (MSVD) -248.00800 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 477.61700 152.72800 Tm (3) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 482.59800 152.72800 Tm [ (\135\054) -249.01700 (Y) 109.98700 (ouCook) -247.99800 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 534.32400 152.72800 Tm (5) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 539.30500 152.72800 Tm (\135\054) Tj -230.44500 -11.95630 Td [ (M\055V) 135.08500 (AD) -445.01900 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 349.02000 140.77200 Tm (32) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 358.98200 140.77200 Tm [ (\135\054) -493.99600 (T) 93.00900 (A) 39.99580 (CoS) -444.99200 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 406.58400 140.77200 Tm (25) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 416.54600 140.77200 Tm (\054) Tj ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 423.47000 140.77200 Tm (28) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 433.43200 140.77200 Tm [ (\135\054) -493.99600 (and) -446 (MPII\055MD) -444.99400 (\133) ] TJ ET Q /R9 cs 0.24707 0.70898 0.28638 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 511.14000 140.77200 Tm (27) Tj ET Q /R11 cs 1 scn q 10 0 0 10 0 0 cm BT /R14 9.96260 Tf 1 0 0 1 521.10300 140.77200 Tm [ (\135\054) -493.99600 (our) ] TJ -212.24300 -11.95510 Td [ (MSR\055VTT) -465.91500 (benchmark) -467.00300 (is) -465.98600 (characterized) -466.99600 (by) -465.99100 (the) -467.01000 (follo) 24.99580 (wing) ] TJ 11.95510 TL T* [ (major) -342.98400 (unique) -342.90100 (properties\056) -588.89900 (First\054) -365.98300 (our) -342.99700 (dataset) -342.90800 (has) -343.11400 (the) -342.91300 (lar) 17.99700 (gest) ] TJ T* [ (number) -268.01400 (of) -267.99200 (clip\055sentence) -266.99700 (pairs\054) -272.99900 (where) -267.01900 (each) -268.00400 (video) -268.01400 (clip) -268.01900 (is) -267.98000 (an\055) ] TJ T* [ (notated) -401.01200 (with) -399.91100 (multiple) -400.98500 (sentences\056) -762.01500 (This) -399.90900 (can) -400.99700 (lead) -400.00700 (to) -401.02100 (a) -400.99700 (bet\055) ] TJ T* [ (ter) -289.99100 (training) -288.90300 (of) -289.98800 (RNNs) -290.00600 (and) -289.98600 (in) -289 (consequence) -289.98100 (the) -289.98100 (generation) -290.01000 (of) ] TJ -13.74140 -29.88790 Td (1) Tj ET Q Q Q q q 1 1 1 rg /a0 gs 48.40600 786.42200 515.18800 -52.69900 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 14.40000 TL ET 1 1 1 rg n 270 47 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 297 50 Tm (5288) 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