%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 12 0 R ] /Type /Pages /Count 10 >> endobj 2 0 obj << /Title (Appearance\055and\055Relation Networks for Video Classification) /Producer (PyPDF2) /Author (Limin Wang\054 Wei Li\054 Wen Li\054 Luc Van Gool) /Subject (2018 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 14 0 R /Resources << /XObject << /x8 15 0 R /x6 18 0 R /x12 21 0 R /x10 24 0 R >> /ExtGState << /s9 27 0 R /s11 30 0 R /a0 << /CA 1 /ca 1 >> /R44 33 0 R /s5 34 0 R /s7 37 0 R >> /Font << /R49 40 0 R /R47 44 0 R /R45 48 0 R /R51 52 0 R /F2 55 0 R /F1 56 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 57 0 R 58 0 R 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 78 0 R 79 0 R 80 0 R 81 0 R 82 0 R 83 0 R 84 0 R 85 0 R 86 0 R 87 0 R 88 0 R 89 0 R 90 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 91 0 R /Resources << /ColorSpace << /R90 92 0 R >> /ExtGState << /R91 94 0 R /R88 95 0 R >> /Font << /R49 40 0 R /R47 44 0 R /R45 48 0 R /F2 96 0 R /F1 97 0 R /R92 98 0 R /R94 102 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 106 0 R 107 0 R 108 0 R 109 0 R 110 0 R 111 0 R 112 0 R 113 0 R 114 0 R 115 0 R 116 0 R 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 ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 137 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /F2 138 0 R /R49 40 0 R /R47 44 0 R /R45 48 0 R /F1 139 0 R /R130 140 0 R /R126 143 0 R /R132 147 0 R /R124 151 0 R /R134 154 0 R /R136 157 0 R /R128 161 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 165 0 R 166 0 R 167 0 R 168 0 R 169 0 R 170 0 R 171 0 R 172 0 R 173 0 R 174 0 R 175 0 R 176 0 R 177 0 R 178 0 R 179 0 R 180 0 R 181 0 R 182 0 R 183 0 R 184 0 R 185 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 186 0 R /Resources << /ColorSpace << /R90 92 0 R >> /ExtGState << /R91 94 0 R /R88 95 0 R >> /Font << /R144 187 0 R /R154 191 0 R /R146 194 0 R /R152 198 0 R /R49 40 0 R /R150 203 0 R /R47 44 0 R /R45 48 0 R /R148 208 0 R /F1 211 0 R /R130 140 0 R /R126 143 0 R /R124 151 0 R /R134 154 0 R /R136 157 0 R /F2 212 0 R /R128 161 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 213 0 R 214 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 215 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /F2 216 0 R /R49 40 0 R /R47 44 0 R /R175 217 0 R /R45 48 0 R /R173 221 0 R /R130 140 0 R /R126 143 0 R /R132 147 0 R /F1 224 0 R /R136 157 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 225 0 R 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R 231 0 R 232 0 R 233 0 R 234 0 R 235 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 236 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /R130 140 0 R /R49 40 0 R /R47 44 0 R /R45 48 0 R /F2 237 0 R /R126 143 0 R /F1 238 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 239 0 R 240 0 R 241 0 R 242 0 R 243 0 R 244 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 245 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /R130 140 0 R /R126 143 0 R /F1 246 0 R /R49 40 0 R /R136 157 0 R /R47 44 0 R /R45 48 0 R /F2 247 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 248 0 R 249 0 R 250 0 R 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 263 0 R 264 0 R 265 0 R 266 0 R 267 0 R 268 0 R 269 0 R 270 0 R 271 0 R 272 0 R 273 0 R 274 0 R 275 0 R 276 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 277 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /R130 140 0 R /F2 278 0 R /F1 279 0 R /R49 40 0 R /R136 157 0 R /R47 44 0 R /R175 217 0 R /R45 48 0 R /R173 221 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 280 0 R 281 0 R 282 0 R 283 0 R 284 0 R 285 0 R 286 0 R 287 0 R 288 0 R 289 0 R 290 0 R 291 0 R 292 0 R 293 0 R 294 0 R 295 0 R 296 0 R 297 0 R 298 0 R 299 0 R 300 0 R 301 0 R 302 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 303 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /R47 44 0 R /F2 304 0 R /R45 48 0 R /F1 305 0 R /R49 40 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 306 0 R /Resources << /ExtGState << /R44 33 0 R >> /Font << /R47 44 0 R /F2 307 0 R /F1 308 0 R /R49 40 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 22245 >> stream q q q 0.1 0 0 0.1 0 0 cm /R44 gs 0 g q 10 0 0 10 0 0 cm BT /R45 14.3462 Tf 1 0 0 1 114.212 675.067 Tm [ (A) 25.0059 (ppearance\055and\055Relation) -249.996 (Netw) 9.99455 (orks) -250.013 (f) 24.9923 (or) -249.995 (V) 37.0137 (ideo) -250.003 <436c6173736902636174696f6e> ] TJ /R47 11.9552 Tf 33.752 -37.8582 Td [ (Limin) -249.989 (W) 80 (ang) ] TJ /R47 7.9701 Tf 59.8109 4.33906 Td (1\0542) Tj /R47 11.9552 Tf 37.3602 -4.33906 Td [ (W) 80 (ei) -249.987 (Li) ] TJ /R47 7.9701 Tf 32.577 4.33906 Td (3) Tj /R47 11.9552 Tf 31.3828 -4.33906 Td [ (W) 80 (en) -250.008 (Li) ] TJ /R47 7.9701 Tf 35.2313 4.33906 Td (2) Tj /R47 11.9552 Tf 31.3828 -4.33906 Td [ (Luc) -250.014 (V) 110.996 (an) -250.01 (Gool) ] TJ /R47 7.9701 Tf 67.068 4.33906 Td (2) Tj -349.525 -13.959 Td (1) Tj /R47 11.9552 Tf 4.4832 -4.33789 Td [ (National) -250 (K) 24.9967 (e) 14.9892 (y) -250 (Laboratory) -250.01 (for) -249.995 (No) 14.9872 (v) 14.9851 (el) -249.987 (Softw) 10 (are) -249.987 (T) 70.0164 (echnology) 65.0132 (\054) -250.012 (Nanjing) -250.01 (Uni) 24.9957 (v) 14.9851 (ersity) 64.9887 (\054) -250.01 (China) ] TJ /R47 7.9701 Tf 8.70898 -9.62031 Td (2) Tj /R47 11.9552 Tf 4.48398 -4.33789 Td [ (Computer) -250.014 (V) 59.9906 (ision) -250.011 (Laboratory) 65 (\054) -250.012 (ETH) -249.997 (Zurich\054) -249.987 (Switzerland) ] TJ /R47 7.9701 Tf 291.369 4.33789 Td (3) Tj /R47 11.9552 Tf 4.4832 -4.33789 Td [ (Google) -249.989 (Research) ] TJ /R45 11.9552 Tf -260.785 -41.0469 Td (Abstract) Tj /R49 9.9626 Tf -83.9281 -24.7949 Td [ (Spatiotempor) 14.995 (al) -437.993 (featur) 37 (e) -438.01 (learning) -438.009 (in) -437.993 (videos) -437.991 (is) -437.993 (a) -438 (funda\055) ] TJ -11.9547 -11.9551 Td [ (mental) -310.004 (pr) 44.9839 (oblem) -310.011 (in) -308.995 (computer) -310.005 (vision\056) -488.988 (This) -309.995 (paper) -310.01 (pr) 36.9865 (esents) -310.012 (a) ] TJ 11.9551 TL T* [ (ne) 15.0177 (w) -271 (ar) 36.9852 (c) 15.0128 (hitectur) 37.0036 (e) 9.99343 (\054) -277.01 (termed) -271.001 (as) ] TJ /R47 9.9626 Tf 113.489 0 Td [ (Appearance\055and\055Relation) -270.984 (Net\055) ] TJ -113.489 -11.9551 Td [ (w) 10.0014 (ork) ] TJ /R49 9.9626 Tf 22.9918 0 Td [ (\050ARTNet\051\054) -266 (to) -263.003 (learn) -262.997 (video) -261.987 (r) 37.0196 (epr) 36.9816 (esentation) -263.015 (in) -263.003 (an) -263.005 (end\055to\055) ] TJ -22.9918 -11.9551 Td [ (end) -372.011 (manner) 110.981 (\056) -678.002 (ARTNets) -372.011 (ar) 36.9852 (e) -373.001 (cons) 0.99248 (tructed) -373.014 (by) -372.016 (stac) 20.0114 (king) -372.984 (multi\055) ] TJ 11.9559 TL T* [ (ple) -269.006 (g) 10.0032 (eneric) -269.008 (b) 20.0016 (uilding) -269.011 (bloc) 20.0089 (ks\054) -273.99 (called) -268.998 (as) ] TJ /R47 9.9626 Tf 152.054 0 Td [ (SMAR) 59.9924 (T) ] TJ /R49 9.9626 Tf 33.723 0 Td [ (\054) -269.011 (whose) -268.989 (goal) ] TJ -185.777 -11.9551 Td [ (is) -378.002 (to) -378.003 (simultaneously) -377.014 (model) -378 (appear) 15 (ance) -378.013 (and) -378 (r) 37.0183 (elation) -377.983 (fr) 44.9851 (om) ] TJ 11.9547 TL T* [ (RGB) -302.018 (input) -303.018 (in) -302.016 (a) -301.984 (separ) 14.9963 (ate) -303 (and) -302.013 (e) 19.9918 (xplicit) -302.984 (manner) 110.981 (\056) -467.018 <53706563690263616c6c79> 54.9957 (\054) ] TJ T* [ (SMART) -292 (bloc) 20.0095 (ks) -291.989 (decouple) -290.996 (the) -291.983 (spatiotempor) 14.995 (al) -292.017 (learning) -291.993 (mod\055) ] TJ T* [ (ule) -312.999 (into) -312.002 (an) -312.997 (appear) 15 (ance) -313.005 (br) 14.9889 (anc) 14.984 (h) -313.002 (for) -312.008 (spatial) -313.004 (modeling) -313.014 (and) -312.992 (a) ] TJ T* [ (r) 37.0196 (elation) -412.016 (br) 14.9889 (anc) 14.984 (h) -412.005 (for) -411.011 (tempor) 15 (al) -411.997 (modeling) 15.0098 (\056) -795.982 (The) -411.995 (appear) 15 (ance) ] TJ 11.9559 TL T* [ (br) 14.9889 (anc) 14.9834 (h) -315.001 (is) -316.012 (implemented) -315.016 (based) -316.009 (on) -314.996 (the) ] TJ /R47 9.9626 Tf 149.716 0 Td [ (linear) -314.999 (combination) ] TJ /R49 9.9626 Tf 78.7828 0 Td (of) Tj -228.499 -11.9551 Td [ (pixels) -328.007 (or) -326.993 <026c746572> -328.003 (r) 37.0196 (esponses) -327.019 (in) -328.011 (eac) 15.0134 (h) -328.019 (fr) 14.9914 (ame) 10 (\054) -346.998 (while) -327.014 (the) -328.016 (r) 37.0183 (elation) ] TJ 11.9551 TL T* [ (br) 14.9889 (anc) 14.9834 (h) -279.007 (is) -278.999 (designed) -278.008 (based) -278.995 (on) -279.002 (the) ] TJ /R47 9.9626 Tf 132.034 0 Td [ (multiplicati) 25 (v) 14.9828 (e) -279.017 (interactions) ] TJ /R49 9.9626 Tf -132.034 -11.9547 Td [ (between) -373.999 (pixels) -373.999 (or) -374.005 <026c746572> -373.995 (r) 37.0196 (esponses) -373.992 (acr) 45.0194 (oss) -373.999 (multiple) -374.018 (fr) 14.9901 (ames\056) ] TJ T* [ (W) 91.9865 (e) -244.003 (perform) -244.01 (e) 19.9918 (xperiments) -242.991 (on) -243.989 (thr) 36.9926 (ee) -244.007 (action) -244.013 (r) 37.0183 (eco) 9.99588 (gnition) -243.996 (benc) 15.0196 (h\055) ] TJ 11.9563 TL T* [ (marks\072) -590.987 (Kinetics\054) -424.988 (UCF101\054) -426.009 (and) -389.999 (HMDB51\054) -425.983 (demonstr) 15.0098 (ating) ] TJ 11.9547 TL T* [ (that) -240.994 (SMART) -240.99 (bloc) 20.0089 (ks) -241.017 (obtain) -240.984 (an) -241.009 (e) 15.0122 (vident) -241.003 (impr) 44.9949 (o) 10.0032 (vement) -241.016 (o) 10.0032 (ver) -240.982 (3D) ] TJ T* [ (con) 39.9988 (volutions) -414.005 (for) -413.991 (spatiotempor) 14.9963 (al) -415.017 (featur) 37.0012 (e) -414.014 (learning) 15.0147 (\056) -803.003 (Under) ] TJ T* [ (the) -219.014 (same) -219.012 (tr) 14.9914 (aining) -218.988 (setting) 9.9922 (\054) -225.019 (ARTNets) -219.998 (ac) 15.0183 (hie) 14.9852 (ve) -218.993 (superior) -218.986 (perfor) 19.982 (\055) ] TJ T* [ (mance) -200.982 (on) -201.015 (these) -201.015 (thr) 36.9926 (ee) -200.014 (datasets) -200.989 (to) -201.013 (the) -201.018 (e) 19.9918 (xisting) -200.991 (state\055of\055the\055art) ] TJ (methods\056) ' ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R47 6.9738 Tf 1 0 0 1 88.298 296.008 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R45 11.9552 Tf 1 0 0 1 50.1121 253.873 Tm [ (1\056) -249.99 (Intr) 18.0146 (oduction) ] TJ /R47 9.9626 Tf 11.9547 -19.8137 Td [ (Deep) -265.995 (learning) -266.997 (has) -265.987 (witnessed) -266.984 (a) -266 (series) -267.007 (of) -265.992 (remarkable) -267.009 (suc\055) ] TJ -11.9547 -11.9551 Td [ (cesses) -475.993 (in) -477.006 (computer) -476.016 (vision\056) -989.013 (In) -475.998 (particular) 40.0056 (\054) -533.007 (Con) 40.0154 (v) 20.0016 (olutional) ] TJ 11.9563 TL T* [ (Neural) -285.008 (Netw) 10.0069 (orks) -286.018 (\050CNNs\051) -284.993 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 159.243 210.148 Tm (25) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 169.205 210.148 Tm [ (\135) -285.995 (ha) 19.9979 (v) 14.9828 (e) -285.016 (turned) -284.997 (out) -286.014 (to) -285 (be) -285.991 (ef) 25.0081 (fec\055) ] TJ -119.093 -11.9547 Td [ (ti) 24.9903 (v) 14.9828 (e) -256 (for) -255.998 (visual) -256.01 (tasks) -256.016 (in) -255.984 (image) -255.981 (domain\054) -258.002 (such) -255.984 (as) -255.994 (image) -255.982 (classi\055) ] TJ 11.9551 TL T* [ <02636174696f6e> -224.011 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 85.5492 186.238 Tm (21) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 95.5117 186.238 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 100.224 186.238 Tm (13) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 110.187 186.238 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 114.909 186.238 Tm (36) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 124.871 186.238 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 129.584 186.238 Tm (39) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 139.546 186.238 Tm [ (\135\054) -228.983 (object) -223.986 (detection) -222.997 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 216.278 186.238 Tm (12) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 226.241 186.238 Tm [ (\135\054) -228.981 (and) -223.997 (semantic) ] TJ -176.129 -11.9551 Td [ (se) 15.0189 (gmentation) -255.01 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 108.951 174.283 Tm (27) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 118.914 174.283 Tm [ (\135\056) -323.986 (Deep) -253.998 (models) -254.986 (ha) 19.9979 (v) 14.9828 (e) -254.002 (been) -255.014 (also) -255.002 (introduced) ] TJ -68.8016 -11.9551 Td [ (into) -268.989 (video) -268.996 (domain) -268.003 (for) -269.015 (action) -268.998 (recognition) -269.019 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 218.371 162.328 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 223.352 162.328 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 228.513 162.328 Tm (46) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 238.475 162.328 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 243.646 162.328 Tm (50) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 253.608 162.328 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 258.779 162.328 Tm (35) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 268.741 162.328 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 273.912 162.328 Tm (49) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 283.875 162.328 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 50.1121 150.373 Tm (41) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 60.0746 150.373 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 64.9266 150.373 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 74.8891 150.373 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 79.741 150.373 Tm (51) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 89.7035 150.373 Tm [ (\135\054) -240 (and) -237.013 (obtain) -236.995 (comparable) -237.006 (or) -237.018 (better) -237.012 (recognition) -236.985 (ac\055) ] TJ -39.5914 -11.9559 Td [ (curac) 15.0061 (y) -319 (to) -318.993 (those) -318.987 (traditional) -319.006 (methods) -319.99 (wi) 0.99003 (th) -320.013 (hand\055crafted) -319.006 (rep\055) ] TJ 11.9551 TL T* [ (resentations) -335.01 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 104.359 126.462 Tm (23) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 114.321 126.462 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 120.159 126.462 Tm (44) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 130.121 126.462 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 135.95 126.462 Tm (47) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 145.912 126.462 Tm [ (\135\056) -567.005 (Ho) 24.9836 (we) 25.0142 (v) 14.9828 (er) 39.986 (\054) -356.017 (the) -336.015 (progress) -334.986 (of) -335.981 (archi\055) ] TJ -95.8 -11.9551 Td [ (tecture) -285.008 (design) -284.987 (and) -285.005 (representation) -284.988 (learning) -284.994 (in) -286.018 (video) -284.994 (domain) ] TJ 11.9547 TL T* [ (is) -251.984 (much) -252.019 (slo) 24.9946 (wer) 39.9884 (\054) -251.995 (partially) -252.009 (due) -251.991 (to) -251.985 (its) -252.016 (inherent) -252.019 (comple) 14.9975 (xity) -252.012 (and) ] TJ ET Q 3.98 w 0 G 501.121 906.789 m 1446.11 906.789 l S q 10 0 0 10 0 0 cm BT /R47 5.9776 Tf 1 0 0 1 60.9711 83.8129 Tm (1) Tj /R47 7.9701 Tf 3.48672 -2.81289 Td [ (The) -249.988 (code) -250.02 (is) -250.001 (at) ] TJ ET Q 0 1 0 0 k q 10 0 0 10 0 0 cm BT /R51 7.9701 Tf 1 0 0 1 110.939 81 Tm (https\072\057\057github\056com\057wanglimin\057ARTNet) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 7.9701 Tf 1 0 0 1 278.312 81 Tm (\056) Tj /R47 9.9626 Tf 30.55 487.246 Td [ (higher) -387.999 (dimension\056) -723.985 (V) 59.9924 (ideo) -387.999 (could) -387.994 (be) -388.014 (vie) 24.986 (wed) -388.989 (as) -388.014 (the) -388.004 (tempo\055) ] TJ 11.9551 TL T* [ (ral) -283.012 (e) 25.0105 (v) 20.0016 (olution) -283.009 (of) -283.009 (a) -283.997 (sequence) -283.007 (of) -283.012 (static) -282.992 (images\056) -409.005 (It) -283.987 (is) -282.997 (generally) ] TJ T* [ (assumed) -359.014 (that) -358.004 (tw) 10.0081 (o) -358.994 (visual) -359.011 (cues) -358.999 (are) -357.989 (crucial) -358.994 (for) -358.999 (video) -358.979 (classi\055) ] TJ T* [ <02636174696f6e> -357.989 (and) -358.994 (understanding\072) -526.02 (\0501\051) ] TJ /R49 9.9626 Tf 131.065 0 Td [ (static) -357.989 (appear) 14.9975 (ance) ] TJ /R47 9.9626 Tf 75.6129 0 Td [ (in) -358.009 (each) ] TJ -206.678 -11.9551 Td [ (frame\054) -464.001 (and) -420.984 (\0502\051) ] TJ /R49 9.9626 Tf 64.7438 0 Td [ (tempor) 15 (al) -421.016 (r) 37.0183 (elation) ] TJ /R47 9.9626 Tf 75.4023 0 Td [ (across) -420.984 (multiple) -422.003 (fra) 1.00964 (mes\056) ] TJ -140.146 -11.9547 Td [ (Therefore\054) -323.012 (an) -307.988 (ef) 25.0081 (fecti) 25.0179 (v) 14.9828 (e) -309.012 (deep) -307.988 (architecture) -309.002 (should) -307.993 (be) -307.988 (able) -309.002 (to) ] TJ 11.9563 TL T* [ (capture) -277.013 (both) -276.991 (information) -277.008 (to) -277.003 (achie) 25.0154 (v) 14.9828 (e) -277.017 (e) 15.0122 (xcellent) -276.993 <636c6173736902636174696f6e> ] TJ 11.9547 TL T* [ (accurac) 14.9975 (y) 65.0137 (\056) ] TJ 11.9551 -16.4754 Td [ (There) -559.007 (are) -557.998 (three) -559.005 (kinds) -557.995 (of) -559.005 (successful) -558.005 (architectures) -558.995 (or) ] TJ -11.9551 -11.9547 Td [ (frame) 25.013 (w) 10 (orks) -493.004 (for) -493.998 (video) -492.999 <636c6173736902636174696f6e> -493.986 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 464.856 456.13 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 469.838 456.13 Tm [ (\135\072) -795.98 (\0501\051) -493.996 (tw) 10.0081 (o\055stream) ] TJ -160.975 -11.9551 Td [ (CNNs) -282.987 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 339.905 444.175 Tm (35) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 349.868 444.175 Tm [ (\135\054) -291.99 (\0502\051) -283.014 (3D) -283.984 (CNNs) -282.987 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 419.068 444.175 Tm (18) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 429.03 444.175 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 434.35 444.175 Tm (41) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 444.313 444.175 Tm [ (\135\054) -291.01 (and) -283.987 (\0503\051) -283.012 (2D) -283.002 (CNNs) -284.007 (with) ] TJ -135.451 -11.9551 Td [ (temporal) -377.991 (models) -379.003 (on) -378.006 (top) -379.018 (such) -378.006 (as) -378.015 (LSTM) -379 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 478.595 432.22 Tm (5) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 483.577 432.22 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 489.833 432.22 Tm (33) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 499.796 432.22 Tm [ (\135\054) -410.99 (temporal) ] TJ -190.934 -11.9551 Td [ (con) 39.9982 (v) 20.0016 (olution) -198.004 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 361.156 420.265 Tm (33) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 371.118 420.265 Tm [ (\135\054) -208.985 (sparse) -198.001 (sampling) -198.001 (and) -198.982 (aggre) 15.0171 (g) 4.98446 (ation) -197.987 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 512.876 420.265 Tm (49) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 522.838 420.265 Tm [ (\135\054) -208.985 (and) ] TJ -213.976 -11.9559 Td [ (attention) -307.993 (modeling) -309.005 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 390.834 408.309 Tm (48) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 400.797 408.309 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 406.356 408.309 Tm (28) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 416.318 408.309 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 421.877 408.309 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 426.859 408.309 Tm [ (\135\056) -485.017 (T) 79.9916 (w) 10.0032 (o\055stream) -308.012 (CNNs) -308.983 (capture) ] TJ -117.996 -11.9551 Td [ (appearance) -286.006 (and) -285.987 (motion) -285.989 (information) -285.987 (with) -286.006 (dif) 24.986 (ferent) -285.996 (streams\054) ] TJ 11.9547 TL T* [ (which) -211.989 (turn) -213.009 (out) -211.987 (to) -211.992 (be) -212.984 (ef) 25.0081 (fecti) 25.0154 (v) 14.9828 (e) -212.009 (for) -212.004 (video) -213.004 (c) 1.01454 <6c6173736902636174696f6e2e> -298.014 (Y) 99.9882 (et\054) -219.998 (it) ] TJ T* [ (is) -237.987 (time\055consuming) -238.012 (to) -237.987 (t) 0.98758 (rain) -237.995 (tw) 10.0081 (o) -237.995 (netw) 10.0081 (orks) -237.985 (and) -237.995 (calculate) -237.995 (opti\055) ] TJ T* [ (cal) -244 <036f> 24.9958 (w) -245.015 (in) -243.986 (adv) 24.9811 (ance\056) -307.993 (T) 79.9916 (o) -245.013 (o) 14.9828 (v) 14.9828 (ercome) -243.999 (this) -244.993 (limitation\054) -244.993 (3D) -245.008 (CNNs) ] TJ 11.9559 TL T* [ (emplo) 9.98363 (y) -327 (3D) -326.997 (con) 39.9982 (v) 20.0016 (olutions) -328.016 (and) -327 (3D) -326.994 (pooling) -327.004 (operations) -327.014 (to) -328.014 (di\055) ] TJ 11.9551 TL T* [ (rectly) -249.99 (learn) -249.997 (spatiotemporal) -249.007 (features) -249.997 (from) -249.988 (stack) 10.0081 (ed) -249.997 (RGB) -250.007 (v) 20.0016 (ol\055) ] TJ T* [ (umes\056) -304.006 (Ho) 24.9836 (we) 25.013 (v) 14.9828 (er) 39.986 (\054) -235.997 (the) -231.993 (performance) -232.981 (of) -232 (3D) -233.01 (CNNs) -232.015 (is) -231.986 (still) -232.986 (w) 10.0032 (orse) ] TJ T* [ (than) -328.011 (tw) 10.0081 (o\055stream) -328.989 (CNNs\054) -347.993 (and) -328.019 (it) -328.989 (is) -328.009 (still) -329.009 (unclear) -327.984 (whether) -328.994 (this) ] TJ T* [ (straightforw) 9.99098 (ard) -434.989 (3D) -435.018 (e) 15.0122 (xtension) -434.991 (o) 14.9828 (v) 14.9828 (er) -434.996 (2D) -435.016 (con) 39.9982 (v) 20.0016 (olution) -434.986 (could) ] TJ T* [ (ef) 25.0081 <026369656e746c79> -378.998 (model) -377.991 (static) -379.015 (appearance) -377.991 (and) -378.991 (temporal) -379.01 (relation\056) ] TJ 11.9563 TL T* [ (2D) -241.991 (CNNs) -242.014 (with) -242.996 (temporal) -242.014 (models) -242.009 (usually) -242.009 (focus) -241.989 (on) -243.008 (capturing) ] TJ 11.9547 TL T* [ (coarser) -268.002 (and) -267.99 (long\055term) -266.982 (temporal) -268.009 (structure\054) -272.013 (b) 20.0016 (ut) -267.985 (lack) -267.99 (capacity) ] TJ T* [ (of) -263.012 (representing) -264.012 <026e6572> -263.003 (temporal) -262.99 (relation) -264.015 (in) -263.005 (a) -263.02 (local) -263.985 (spatiotem\055) ] TJ T* [ (poral) -249.988 (windo) 25 (w) 65.0112 (\056) ] TJ 11.9551 -16.475 Td [ (In) -352.017 (this) -353.015 (paper) 39.9909 (\054) -378.013 (we) -351.983 (address) -352.99 (the) -352.01 (problem) -353.015 (of) -352.02 (capturing) -353.015 (ap\055) ] TJ -11.9551 -11.9551 Td [ (pearance) -258.011 (and) -259.011 (relation) -258.016 (in) -257.984 (video) -258.016 (domain\054) -260.981 (by) -257.986 (proposing) -257.996 (a) -258.981 (ne) 25.0154 (w) ] TJ T* [ (architecture) -416.007 (unit) -415.984 (termed) -415.997 (as) ] TJ /R49 9.9626 Tf 115.071 0 Td (SMART) Tj /R47 9.9626 Tf 35.1379 0 Td [ (block\056) -808.002 (Our) -416.004 (SMAR) 59.99 (T) ] TJ -150.209 -11.9559 Td [ (block) -332.986 (aims) -332 (to) -332.991 (Simultaneously) -331.984 (Model) -333.018 (Appearance) -331.989 (and) -332.998 (Re\055) ] TJ T* [ (laT) 34.9819 (ion) -272.016 (from) -271.984 (RGB) -272.004 (input) -272.004 (in) -271.984 (a) -271.999 (separate) -271.994 (and) -273.008 (e) 15.0122 (xplicit) -271.999 (w) 10.0032 (ay) -271.994 (with) ] TJ T* [ (a) -311.012 (tw) 10.0081 (o\055branch) -311.982 (unit\054) -327.017 (in) -310.995 (contrast) -312.007 (to) -310.997 (modeling) -311.982 (them) -310.987 (with) -312.002 (tw) 10.0081 (o\055) ] TJ T* [ (stream) -294.015 (inputs) -293.015 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 368.947 152.731 Tm (35) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 378.909 152.731 Tm [ (\135) -293.993 (or) -293.988 (jointly) -293.005 (and) -293.985 (implicitly) -294.01 (with) -292.985 (a) -293.995 (3D) -293.98 (con\055) ] TJ -70.0473 -11.9551 Td [ (v) 20.0016 (olution) -374.014 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 348.922 140.776 Tm (41) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 358.884 140.776 Tm [ (\135\056) -682.984 (As) -374.006 (sho) 24.9934 (wn) -374.006 (in) -374.986 (Figure) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 457.006 140.776 Tm (1) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R47 9.9626 Tf 1 0 0 1 461.987 140.776 Tm [ (\054) -404.986 (our) -374.991 (SMAR) 59.99 (T) -374.011 (block) ] TJ -153.125 -11.9559 Td [ (is) -394 (a) -394.998 (multi\055branch) -394.013 (architecture\054) -430.002 (which) -395.017 (is) -393.998 (composed) -393.983 (of) -394.993 (ap\055) ] TJ 11.9551 TL T* [ (pearance) -333.998 (branch) -333.981 (and) -334.018 (relation) -334.003 (branch\054) -354.995 (and) -334.018 (fuses) -334.013 (them) -334.003 (with) ] TJ T* [ (a) -357.004 (concatenation) -357.014 (and) -356.994 (reduction) -357.014 (operation\056) -631.007 (The) -356.994 (appearance) ] TJ T* [ (branch) -251.014 (is) -251.005 (based) -251.01 (on) -251.007 (t) 0.98758 (he) ] TJ /R49 9.9626 Tf 91.0719 0 Td [ (linear) -250.992 (combination) ] TJ /R47 9.9626 Tf 78.6121 0 Td [ (of) -251.017 (pix) 14.9975 (els) -251.007 (or) -251.017 <026c746572> ] TJ -169.684 -11.9551 Td [ (responses) -199.004 (in) -199.013 (each) -198.994 (frame) -198.989 (to) -199.016 (model) -199.001 (spatial) -199.001 (structure\054) -209.985 (while) -199.006 (the) ] TJ -13.741 -29.8879 Td (1) 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 47 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 50 Tm (1430) Tj T* ET Q endstream endobj 15 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 16 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 16 0 obj << /Filter /FlateDecode /Resources 17 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
17 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
18 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 19 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
19 0 obj
<<
/Filter /FlateDecode
/Resources 20 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
20 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 22 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
22 0 obj
<<
/Filter /FlateDecode
/Resources 23 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