Check It Again:论文整理
蕴含简介
- 文本蕴含:文本间的推理关系,又称为文本蕴含关系,作为一种基本的文本间语义联系,广泛存在于自然语言文本中。简单的来说文本蕴含关系描述的是两个文本之间的推理关系,其中一个文本作为前提,另一个文本作为假设,如果根据前提P能够推理出假设H,那么就说P蕴含H,记作P->H(前提->假设)。
- 视觉蕴含:也就是把前提改成了图片,而不是之前的文字,由图片来推出假设是否成立。
解决问题
作者发现,目前通用的VQA模型,或是基于语言先验的模型,都产生了一个不可忽略的问题。都通过一个最佳输出来预测正确答案,而不检查答案的真实性。而且这些模型没有很好的利用答案的语义信息来缓解语言先验。
为了解决上述问题,提出了本文的方法。
研究方法和创新点
作者提出了一个选择再排序的渐进框架来解决语言先验问题。表明这个框架可以很好的和其他模型结合,并进一步提高他们的能力。强调了文本和图像之间的验证过程,并将VQA任务表述为一个视觉蕴含问题。该过程充分利用了图像、问题和候选答案的交互信息。
选择重新排序(SAR)框架:由一个候选答案选择模块(Candidate Answer Selecting module)和一个答案重新排序模块Answer Re-ranking module组成。
CAS:给定一个图像和一个问题,我们首先使用当前VQA模型得到一个前N个答案组成的候选答案集。这个模块可以将无关的答案过滤掉。
VE:在答案重新排序模块中将VQA表示为VE任务,其中图像的前提,合成的答案和问题的组合是假设。
流程和模块:
SAR
LXMERT
图像特征包含位置信息
- CAS
CAS首先给出所有可选答案的回归分数,然后CAS选择N个得分最高的答案作为候选,即为:
- Answer Re-ranking module
一个问题和每个候选答案可以桥接成一个完整的陈述,然后图像可以验证每个陈述的真实性。将VQA表述为一个VE问题,其中图像是前提,合成陈述是假设。合成语句也称为”密集标题“。- 问答组合策略
- R 用答案替换类别前缀。如:No a crosswalk是Is this a crosswalk的结果。
- C 直接把问题和答案串联起来。例如:8 How many flowers in the case?
- R→C 在训练时使用R策略,防止模型过度关注问题类别和答案的共现关系,测试时使用C策略,引入更多信息进行推断。
- 问答组合策略
- 使用预训练的LXMERT对图像和答案问题组合的视觉蕴含度进行评分。将图像及其第n个候选标题的VE评分表示为:
Trm()是LXMERT之后密集层的输出。分数越大则表示蕴含度越高。通过最小化多标签软损失来优化参数:
- 结合语言先验法
实验设置
- 数据集:VQA-CP VQA v2
- baselines:
- UpDn,Areg,…
- LXMERT
- SAR+SSL
- epoch 20
- batch size 32
- SAR+LMH
- epoch 10
- batch size 64
在用自监督损失微调模型之前,我们不需要用VQA损失预先训练模型。
- learning rate 1e-5
- N
实验结果和分析
模型朝着真正全面的VQA模型取得了重大进展。与以往的方法不同的是,我们的方法可以减少语言先验,同时保持良好的回答问题的能力。
随着N的增加,Accuracy逐渐变好。作者认为增加N,可以进一步获得更好的性能。
作者发现选择更好的VQA模型作为CAS并不能保证更好的性能。
当N太小时,正确答案可能无法被CAS召回;当N过大时,错误答案的影响使模型更难选择正确答案。
结论
基于图像蕴含的SAR渐进框架,建立了一个新的最先进的准确率为66.73%。
论文代码
论文图像补充
top12+SAR
Top20-SAR+LMH
n对不同问题类型的影响
lmh:
ssl:
答案分布
'living room':0 'skateboarding':1 '7':2 'cake':3 'outside':4 'air':5 'daytime':6 'motorcycles':7 'cat':8 'flags':9
'england':10 'ocean':11 'pottery':12 'on rock':13 'stickers':14 'coat':15 'parasailing':16 'skiing':17 'cart':18 'mushroom':19
'bathroom':20 'wine glass':21 'bright':22 'side':23 'hotel':24 'collar':25 'all':26 '6 feet':27 'stuffed animals':28 'straight':29
'15':30 'flying kite':31 'batting':32 'to catch ball':33 'curtains':34 'broken':35 'barber shop':36 'keyboard':37 'horse racing':38 'sleep':39
'button up':40 'throw':41 'german shepherd':42 'paint':43 'picnic table':44 'catcher':45 'overalls':46 'fountain':47 'tennis':48 'land':49
'stir fry':50 'leather':51 'football':52 'background':53 'wooden':54 'warm':55 'sad':56 'sugar':57 'teddy bear':58 'chain link':59
'cookies':60 'forks':61 'spoons':62 'visor':63 'team':64 'talking on phone':65 'kids':66 'blender':67 'cactus':68 'orange and yellow':69
'rough':70 'linoleum':71 'selfie':72 'parsley':73 'red white and blue':74 'saddle':75 'cell phone':76 'vegetarian':77 'gas station':78 'water bottle':79
'fish':80 'railing':81 'plain':82 'male':83 'nose':84 'cheese':85 'surfboards':86 'heavy':87 'lemon':88 'space needle':89
'dalmatian':90 'ducati':91 'business':92 'sidewalk':93 'shell':94 'gravel':95 'clouds':96 'beef':97 'cloth':98 'floor':99
'watching tv':100 'skateboarder':101 'olives':102 'kitchen':103 'finch':104 'cardboard':105 'bookshelf':106 'rooster':107 'female':108 'chili':109
'boxes':110 'half full':111 'toothbrush':112 'blood':113 'bicycles':114 'blinders':115 'china':116 'ford':117 'kid':118 'not sure':119
'afternoon':120 'hammer time':121 'leaving':122 'walking':123 'sandwiches':124 'single':125 'theater':126 'eggs':127 'carpet':128 'cotton':129
'1 world':130 'skis':131 'fern':132 'macaroni':133 'cupcake':134 'fridge':135 'bat':136 'plates':137 'driver':138 'older':139
'handle':140 'magnets':141 'cereal':142 'helmet':143 'logo':144 'deep':145 'horses':146 'playing soccer':147 'checkered':148 'landing':149
'bleachers':150 'snow':151 'apple and banana':152 'umpire':153 'fly kite':154 '10:20':155 'screen':156 'before':157 'toy':158 'sand':159
'lamb':160 'indian':161 'beer':162 'polar':163 'reading':164 'on chair':165 'tattoo':166 'wine tasting':167 'rocks':168 'bedroom':169
'green':170 'los angeles':171 'coins':172 'yamaha':173 '12':174 'shorts':175 'tropicana':176 'horseback riding':177 'steak':178 'work':179
'yes':180 'race':181 'metal':182 'setting':183 'magazine':184 'caucasian':185 'jockey':186 'sled':187 'earrings':188 'sushi':189
'circus':190 'stopped':191 '2 feet':192 'dessert':193 'heineken':194 'poles':195 'ham':196 'skating':197 'street':198 'tiger':199
'la':200 'starbucks':201 'fire hydrant':202 'seat':203 'alive':204 '17':205 'buildings':206 'tape':207 'licking':208 'paddle':209
'parrot':210 'pavement':211 'apron':212 'transportation':213 'racket':214 'sunflowers':215 'pictures':216 'headband':217 'shearing':218 'cutting board':219
'on bike':220 'wall':221 'birds':222 'formica':223 'peeing':224 'spinach':225 'fake':226 'serve':227 'passengers':228 'young':229
'military':230 'roses':231 'toothpaste':232 'cream':233 'fighting':234 'black':235 'chopsticks':236 'resting':237 'chiquita':238 'snowboarding':239
'red':240 'desktop':241 'new':242 'meter':243 'crane':244 'coke':245 'pm':246 'fanta':247 'fried':248 'camping':249
'maroon':250 '21':251 'doll':252 'crib':253 'cleaner':254 'soldier':255 'glazed':256 'styrofoam':257 '60':258 'cargo':259
'rowing':260 'outdoors':261 'concrete':262 'pitcher':263 'jet ski':264 'surfboard':265 'dodge':266 'mountains':267 'google':268 'public':269
'lady':270 'using computer':271 'tongue':272 'bagel':273 'countryside':274 'fork and knife':275 'maple':276 'electricity':277 'trucks':278 'snowboard':279
'sweet':280 'multi colored':281 'serving':282 'pole':283 'kitten':284 'wicker':285 'hair':286 '39':287 'multicolored':288 '14':289
'hay':290 'lays':291 'happy':292 'statues':293 'cleats':294 '34':295 'formal':296 'mets':297 'waiting':298 'overcast':299
'power lines':300 'raw':301 'weather vane':302 'north':303 'meow':304 'bmw':305 'standing':306 '47':307 'jp morgan':308 'boating':309
'wide':310 'shaking hands':311 'robe':312 'decorative':313 'granite':314 'poodle':315 'goggles':316 'herd':317 'dragon':318 'ceramic':319
'wii remotes':320 'house':321 'movie':322 'branch':323 'cross country':324 'cabinet':325 '2012':326 'swan':327 'working':328 'breakfast':329
'talking':330 'mouse pad':331 'junk':332 'volkswagen':333 'planes':334 'fruit':335 'mustache':336 'pedestal':337 'classroom':338 'meeting':339
'hiking':340 'snowsuit':341 'driving':342 'restaurant':343 'evening':344 'dress':345 'surfing':346 'in water':347 'swinging':348 'zebra':349
'beige':350 'don't walk':351 'tunnel':352 'orioles':353 'sleeping':354 'airport':355 'wild':356 'on street':357 'dip':358 'square':359
'victorian':360 'sandwich':361 'painting':362 'straw':363 'toys':364 'unsure':365 'night':366 'cafe':367 'tools':368 'wetsuit':369
'tennis player':370 'landscape':371 'focus':372 'parked':373 'polar bears':374 'boys':375 'dirt':376 'clock':377 'celery':378 'electric':379
'texting':380 'crown':381 'family':382 'barrel':383 'unknown':384 'navy':385 'kiwi':386 'knives':387 'glass':388 'laptop':389
'television':390 'leaves':391 'closed':392 'casual':393 'handicap':394 'riding':395 'information':396 'rope':397 'thick':398 'supreme':399
'tag':400 'dry':401 'socks':402 'downhill':403 'gloves':404 '4 way':405 'blueberry':406 'black and white':407 'dirty':408 'dining room':409
'africa':410 'relaxing':411 'cars':412 'plunger':413 'chinese':414 'modern':415 'pier':416 '18':417 'monday':418 'african':419
'snowboarder':420 'bus':421 'crafts':422 'lettuce':423 'orange juice':424 'helmets':425 'arabic':426 'east':427 'cucumber':428 'rice':429
'terrier':430 'comfort':431 '4 inches':432 'tomato':433 'asian':434 'bathtub':435 '1':436 'salmon':437 'goat':438 'urinal':439
'stove':440 'porcelain':441 'pilot':442 '5:50':443 'weeds':444 'beans':445 '30':446 'mcdonald's':447 'color':448 'virgin':449
'neither':450 'uphill':451 'summer':452 'roof':453 'train tracks':454 'referee':455 'winnie pooh':456 'ceiling':457 'donut':458 'rodeo':459
'swimsuit':460 'canopy':461 'germany':462 'tissue':463 '120':464 'photographer':465 'softball':466 'still':467 'baby':468 'adult':469
'9:35':470 'news':471 'design':472 'tabby':473 'pasture':474 'toshiba':475 'grilled':476 'plane':477 'fog':478 'skiers':479
'peas':480 'striped':481 'couch':482 'serious':483 'giraffes':484 'tea':485 'water ski':486 'forward':487 'reflection':488 'peanut butter':489
'clothes':490 '31':491 'pirate':492 'large':493 'blankets':494 'lights':495 'ollie':496 'sun':497 'fur':498 'water skiing':499
'upside down':500 'thanksgiving':501 'windsurfing':502 'sepia':503 'pepper':504 'hawaii':505 'tennis racquet':506 'drain':507 'vase':508 'compaq':509
'dots':510 'boat':511 'relish':512 'turkey':513 'kite':514 'track':515 'tank top':516 'crosswalk':517 'sweatband':518 '3:55':519
'market':520 'triangle':521 'lace':522 'ambulance':523 'real':524 'eiffel tower':525 'sunny':526 'waves':527 'riding bike':528 'gate':529
'towing':530 'jacket':531 'words':532 'peppers':533 'purse':534 'morning':535 'hummingbird':536 'elephants':537 'bowling':538 'skier':539
'winter':540 '53':541 'freight':542 'cookie':543 'moon':544 'bowtie':545 'candle':546 'many':547 'nightstand':548 '2':549
'lines':550 'shirt':551 'victoria':552 '':553 'boogie board':554 'air conditioner':555 'nintendo':556 'computer':557 'pitching':558 '16':559
'11:05':560 'shoe':561 'lake':562 'lab':563 'friend':564 'urban':565 'lexus':566 'pc':567 'balance':568 'downtown':569
'bricks':570 'museum':571 '4:55':572 'can':573 'zoo':574 'round':575 'grass':576 'no':577 'icing':578 'no parking':579
'oranges':580 'bow':581 'river':582 'keys':583 'van':584 'denim':585 'aa':586 'braid':587 'riding motorcycle':588 'north america':589
'cartoon':590 'leaf':591 '11':592 'pancakes':593 'bird feeder':594 'ducks':595 'road':596 '7 eleven':597 'strap':598 'huge':599
'no hat':600 'flamingo':601 'cigarette':602 'traffic light':603 'ferris wheel':604 'males':605 'hawaiian':606 '27':607 'conference':608 '32':609
'suitcases':610 'broccoli':611 'jump':612 'golden retriever':613 'sunglasses':614 'leopard':615 'umbrella':616 'ladder':617 'direction':618 'bathing':619
'clip':620 'frog':621 'fire':622 'cardinals':623 'ground':624 'very long':625 'wool':626 'fries':627 'nike':628 'buoys':629
'posing':630 'mirror':631 'red white blue':632 'on car':633 'lot':634 'uniforms':635 'juice':636 'riding horse':637 'directions':638 'fell':639
'rural':640 'christmas':641 'top hat':642 'pepsi':643 'kites':644 'motorola':645 'geese':646 'stick':647 'tam':648 'h':649
'clear':650 'nothing':651 'commercial':652 'laying down':653 'both':654 'bench':655 'surfers':656 'ping pong':657 'healthy':658 'plants':659
'smiling':660 'orange':661 '40':662 'shore':663 'glasses':664 'umbrellas':665 'typing':666 'old':667 'frosting':668 'heater':669
'metro':670 'blurry':671 'snowing':672 'verizon':673 'right':674 'tie':675 'front':676 'tomatoes':677 'yogurt':678 'smoothie':679
'horizontal':680 'tater tots':681 'hugging':682 'necklace':683 'herding':684 'ski slope':685 'steps':686 'dinner':687 'hamburger':688 'rottweiler':689
'greyhound':690 'napkin':691 'hello kitty':692 'wii remote':693 'tractor':694 'crosstown':695 'coconut':696 'across street':697 'tray':698 'alcohol':699
'hot':700 'blinds':701 'right hand':702 'honda':703 'scarf':704 'wet':705 'pan':706 'in car':707 'lanyard':708 'advertisement':709
'potato':710 '44':711 'sheep':712 'tennis rackets':713 'blue':714 'shaved':715 '100 feet':716 'sunlight':717 '20':718 'give way':719
'pans':720 'shepherd':721 'tank':722 'chef':723 'vest':724 'fly':725 'arm':726 '24':727 'controller':728 'chicken':729
'fluorescent':730 'o':731 'frisbee':732 'foggy':733 'game':734 'magnet':735 'vases':736 'taking picture':737 'alaska':738 'door':739
'rectangle':740 'backpack':741 'school':742 'plaster':743 'statue':744 'woods':745 'vines':746 '1 in front':747 'pine':748 'clock tower':749
'lighthouse':750 'hsbc':751 'hat':752 'pigtails':753 'champagne':754 'warning':755 'earring':756 'chairs':757 'lunch':758 'back':759
'scissors':760 'samsung':761 'spoon':762 'suv':763 'emergency':764 'hp':765 'rock':766 'kite flying':767 'moving':768 'legos':769
'toyota':770 'scooter':771 'white':772 'clocks':773 'giants':774 'thin':775 'rays':776 'tulips':777 'polo':778 'steam':779
'kawasaki':780 'on toilet':781 'butterfly':782 'c':783 'skate park':784 'krispy kreme':785 'desk':786 'blue and orange':787 'ice cream':788 'burger':789
'biking':790 'drinking':791 'smiley face':792 '2000':793 'cooking':794 '2:05':795 'asphalt':796 'shrimp':797 'butter':798 'blanket':799
'curb':800 'books':801 'camper':802 'tower':803 'windowsill':804 'british airways':805 'silver':806 'rose':807 'style':808 'fedex':809
'married':810 'police':811 'they aren't':812 'adidas':813 'grill':814 'fast':815 'limes':816 'billabong':817 'windsor':818 'halloween':819
'computers':820 'seagull':821 '55':822 'head':823 'scania':824 'gray and black':825 'pots':826 'runway':827 'bulldog':828 'rust':829
'store':830 'stained glass':831 'rain':832 'parade':833 '2:10':834 'loading':835 'kayak':836 'coal':837 'coca cola':838 'vanilla':839
'forest':840 'stainless steel':841 'noodles':842 'engine':843 'american':844 'west':845 'soccer':846 'grocery':847 'signs':848 'sprinkles':849
'radish':850 '11:35':851 '1 4':852 'red and gray':853 'tracks':854 'indoors':855 'inside':856 'city bus':857 'palm':858 'ipod':859
'bike':860 'yield':861 'kitesurfing':862 'petting horse':863 'people':864 'honey':865 'soup':866 'tree':867 'storage':868 'open':869
'rug':870 'pear':871 'arizona':872 'drywall':873 'tired':874 'drawing':875 'dresser':876 'toilet paper':877 'knee pads':878 'fork':879
'moss':880 'tablet':881 'laptops':882 'guitar':883 'badminton':884 'cross':885 'below':886 'elephant':887 'skateboard':888 'bananas':889
'left side':890 'wristband':891 'cardinal':892 'cows':893 'sunset':894 'phones':895 'beanie':896 'knife':897 'carriage':898 'backhand':899
'eating':900 'motorcycle':901 'not in service':902 'headphones':903 'doubles':904 'gray and white':905 'free':906 'goal':907 'captivity':908 '2.00':909
'coleslaw':910 'tennis ball':911 'muffin':912 'new york':913 'wreath':914 'toilet':915 'food':916 'high chair':917 'full':918 'stool':919
'clothing':920 'red and green':921 '400':922 'travel':923 'dell':924 'towels':925 'small':926 'hearts':927 'pot':928 'menu':929
'skyscraper':930 'city':931 'emirates':932 'pasta':933 'farmer':934 'no number':935 'shelf':936 'fireplace':937 'bed':938 'bikini':939
'can't tell':940 'penguin':941 'grapefruit':942 'skirt':943 'apartment':944 'dusk':945 'ball':946 'boxing':947 'curly':948 'pointing':949
'red sox':950 'brace':951 'nature':952 'pickle':953 'nokia':954 'man':955 '0':956 'bamboo':957 'peacock':958 'ties':959
'harbor':960 'adults':961 'shadows':962 'man's':963 'suit':964 'bears':965 'apples':966 'paper':967 'ice':968 'braves':969
'ascending':970 'watermelon':971 'flowers':972 'tents':973 'fan':974 'bunk':975 'trash can':976 'red and blue':977 'bird':978 'mozzarella':979
'children':980 'parking meters':981 '25':982 'snowboards':983 'farm':984 'yellow':985 'lion':986 'gray':987 'sneakers':988 'collie':989
'asleep':990 'microwave':991 'wheelchair':992 '6:20':993 'skull and crossbones':994 'mud':995 'squares':996 'birthday':997 'high':998 'pants':999
'climbing':1000 '13':1001 'ride':1002 'monkey':1003 'corn':1004 'bud light':1005 'fence':1006 'boxer':1007 'hands':1008 'wood':1009
'50':1010 'parking':1011 'abstract':1012 'beard':1013 'bread':1014 'yellow and red':1015 'obama':1016 'hair dryer':1017 'dachshund':1018 'chihuahua':1019
'bag':1020 'balloons':1021 'electronics':1022 'basil':1023 'ponytail':1024 'brewers':1025 'baseball':1026 'bandana':1027 'onions':1028 'sweat':1029
'reins':1030 'pink':1031 'unclear':1032 'singles':1033 'kickstand':1034 '6:00':1035 'cleaning':1036 'mutt':1037 'curtain':1038 '9:45':1039
'bottle':1040 '23':1041 'sunflower':1042 'fair':1043 'wireless':1044 'csx':1045 'passenger':1046 '68':1047 '11:55':1048 'taxi':1049
'on counter':1050 'hill':1051 'jet':1052 'sink':1053 'volleyball':1054 'shed':1055 'fall':1056 'ram':1057 '10:55':1058 'selling':1059
'mt airy':1060 't shirt':1061 'strawberry':1062 'stop':1063 'safety':1064 'cement':1065 'pink and white':1066 'grapes':1067 'go':1068 'ana':1069
'flip flops':1070 'white and brown':1071 'coffee':1072 'bush':1073 'propeller':1074 'santa':1075 'speakers':1076 'french fries':1077 'sprite':1078 'microphone':1079
'chain':1080 'puma':1081 'deer':1082 'building':1083 'toothpick':1084 'giraffe':1085 'stairs':1086 'spotted':1087 'caution':1088 'delivery':1089
'yarn':1090 'tennis court':1091 'deck':1092 'on motorcycle':1093 'towards':1094 'houses':1095 'converse':1096 'bun':1097 'basketball':1098 'coats':1099
'arriving':1100 'wind':1101 'no dog':1102 'peace':1103 'side of road':1104 'string':1105 '5:15':1106 'kettle':1107 'usa':1108 'tan':1109
'multi':1110 'concert':1111 'flat':1112 'mask':1113 'mexican':1114 'harness':1115 'copper':1116 'home':1117 'no water':1118 'oven':1119
'wine glasses':1120 'toaster oven':1121 'left':1122 'clay':1123 'bridle':1124 'human':1125 'potatoes':1126 'shopping cart':1127 'fire truck':1128 'gothic':1129
'rugby':1130 'hockey':1131 'logs':1132 'hood':1133 'cherry':1134 'sedan':1135 'away':1136 'cumulus':1137 'blackberry':1138 'no clock':1139
'bucket':1140 'don't know':1141 'warmth':1142 'sliding':1143 'disney':1144 'cats':1145 'pull':1146 'forehand':1147 'wallet':1148 'down':1149
'almonds':1150 'milk':1151 'casserole':1152 'cowboy':1153 '4':1154 'airplane':1155 'motor':1156 'toilets':1157 'pig':1158 'pool':1159
'buoy':1160 'mac':1161 '200':1162 'plaid':1163 'orange and white':1164 'wallpaper':1165 'long':1166 'cane':1167 'nuts':1168 'smoke':1169
'bracelet':1170 'on phone':1171 'benches':1172 'up':1173 'goatee':1174 'dog':1175 'boats':1176 'oars':1177 'swimming':1178 'light':1179
'newspaper':1180 'tusks':1181 'cloudy':1182 'earbuds':1183 'bridge':1184 'outdoor':1185 'tuna':1186 'out':1187 'd':1188 'town':1189
'humans':1190 'pastries':1191 'iphone':1192 'candles':1193 'tent':1194 'camouflage':1195 'plant':1196 'mustard':1197 'dunkin donuts':1198 'wii controllers':1199
'couple':1200 'photography':1201 'bull':1202 'baskets':1203 'beagle':1204 'mountain':1205 'cherries':1206 'cuddling':1207 'sandals':1208 'brush':1209
'desert':1210 'red and black':1211 'foam':1212 'running':1213 'tugboat':1214 'money':1215 'buses':1216 'playing':1217 '26':1218 'behind woman':1219
'2:20':1220 'on shelf':1221 'watching':1222 'camel':1223 'frame':1224 'watch':1225 'gatorade':1226 'blue and red':1227 'solid':1228 'lufthansa':1229
'german':1230 'goats':1231 'domestic':1232 'woman':1233 'steamed':1234 'tire':1235 '6:05':1236 'soldiers':1237 'church':1238 '30 mph':1239
'triangles':1240 'smartphone':1241 'plate':1242 'olympics':1243 'highway':1244 'left and right':1245 'box':1246 'daisies':1247 'lamp':1248 'sail':1249
'europe':1250 'no 1':1251 'dugout':1252 'apartments':1253 'hand':1254 'art':1255 'globe':1256 'ski poles':1257 'asics':1258 'suitcase':1259
'w':1260 '10:35':1261 '4:40':1262 'octagon':1263 'lilies':1264 'twins':1265 'sky':1266 'thousands':1267 'italian':1268 'cnn':1269
'vegetables':1270 'table':1271 'opaque':1272 'dancing':1273 'catching frisbee':1274 'girl':1275 'bnsf':1276 'shade':1277 'uk':1278 'sticks':1279
'5:30':1280 'main street':1281 'desserts':1282 'normal':1283 'chicago':1284 'beach':1285 'red and yellow':1286 'flying':1287 'toilet brush':1288 'joshua':1289
'pearl':1290 'remote':1291 'on':1292 'pigeons':1293 'antique':1294 'ostrich':1295 'bowl':1296 'tub':1297 'chair':1298 'multiple':1299
'barn':1300 'ferry':1301 'floral':1302 'mushrooms':1303 'mound':1304 'helicopter':1305 'animals':1306 'cucumbers':1307 'russian':1308 'salad':1309
'leash':1310 'balls':1311 'shadow':1312 '3:00':1313 'lg':1314 'hose':1315 'court':1316 'blue and yellow':1317 'suzuki':1318 'regular':1319
'sitting':1320 'pillows':1321 'papers':1322 'good':1323 'acer':1324 'jeans':1325 '7:35':1326 'picnic':1327 'kia':1328 'truck':1329
'35':1330 'cook':1331 'wii controller':1332 'fishing':1333 'sweatshirt':1334 'analog':1335 'scratching':1336 'to left':1337 'website':1338 'all way':1339
'eagle':1340 'cylinder':1341 'camo':1342 '5:00':1343 'wine':1344 'chevron':1345 'pond':1346 'chase':1347 'g':1348 'curved':1349
'bus stop':1350 'looking':1351 'cold':1352 'chandelier':1353 'playing video game':1354 'playing frisbee':1355 'dishwasher':1356 'stars':1357 'chips':1358 'brown':1359
'belt':1360 'wire':1361 'spray paint':1362 'tarmac':1363 'messy':1364 'parking lot':1365 'office':1366 'jungle':1367 'microsoft':1368 'shower curtain':1369
'traffic':1370 'radio':1371 'jones':1372 'slow down':1373 '1st':1374 'gravy':1375 'gas':1376 'off':1377 '2:50':1378 '15 feet':1379
'brick':1380 'child':1381 'wheat':1382 'circle':1383 'graffiti':1384 'lilac':1385 'america':1386 'shirts':1387 'wagon':1388 'taking off':1389
'cabinets':1390 '5':1391 'night time':1392 'mouse':1393 'natural':1394 'husky':1395 'christmas tree':1396 'mercedes benz':1397 'batter':1398 'play':1399
'11:30':1400 'garage':1401 'fedora':1402 'jeep':1403 'virgin atlantic':1404 'carrots':1405 'pajamas':1406 'sunbathing':1407 'protection':1408 'polka dot':1409
'stones':1410 'fruits':1411 'nissan':1412 'field':1413 'branches':1414 'kenmore':1415 '9:05':1416 'counter':1417 'bottom':1418 'day':1419
'budweiser':1420 'pug':1421 'pork':1422 'ring':1423 'book':1424 'oar':1425 'packing':1426 'bakery':1427 'soda':1428 'high heels':1429
'container':1430 'ski lift':1431 'space':1432 'boy':1433 'bidet':1434 'horse':1435 'clydesdale':1436 'venice':1437 'bus station':1438 'tying tie':1439
'sheet':1440 'banana':1441 'canada':1442 'stripe':1443 'net':1444 'westjet':1445 'no shirt':1446 'short':1447 'person':1448 'biplane':1449
'monitor':1450 'tennis racket':1451 'green and yellow':1452 'pickles':1453 'not possible':1454 'corner':1455 'drinks':1456 'big ben':1457 'walk':1458 'california':1459
'men':1460 '8':1461 'play tennis':1462 'sweater':1463 'banana peel':1464 '3':1465 '46':1466 'on sidewalk':1467 'accident':1468 '9:12':1469
'not':1470 'double decker':1471 'tags':1472 'tongs':1473 'license plate':1474 'recliner':1475 'crest':1476 'knife and fork':1477 'soft':1478 'bicycle':1479
'trash':1480 'gmc':1481 'in':1482 'train station':1483 'pitch':1484 'spices':1485 '22':1486 '56':1487 'pelican':1488 'carrot':1489
'ketchup':1490 'stucco':1491 'wetsuits':1492 'airplanes':1493 '28':1494 'water':1495 'bags':1496 'mercedes':1497 'vans':1498 'black and red':1499
'toast':1500 'playing game':1501 'factory':1502 'owl':1503 'library':1504 'surfer':1505 'right 1':1506 'produce':1507 'on top':1508 '6':1509
'black and gray':1510 'far':1511 'holding it':1512 '2010':1513 'white and red':1514 'bikes':1515 'stop sign':1516 'for sale':1517 'school bus':1518 'atv':1519
'bar':1520 'handlebars':1521 'sports':1522 'ibm':1523 'cutting cake':1524 'easter':1525 'sign':1526 'blue and black':1527 'hardwood':1528 'traffic lights':1529
'car':1530 'sailboat':1531 'shells':1532 'chocolate':1533 'not likely':1534 '1 in middle':1535 'in box':1536 'refrigerator':1537 'in bowl':1538 'baseball bat':1539
'hospital':1540 'glaze':1541 'pens':1542 'cameras':1543 'painted':1544 'calm':1545 'on couch':1546 'oval':1547 'alligator':1548 'petting':1549
'barbed wire':1550 'diamond':1551 'harley davidson':1552 'dairy':1553 'siamese':1554 '1:50':1555 'pigeon':1556 'dirt bike':1557 'wedding':1558 'movement':1559
'taking pictures':1560 'western':1561 'steel':1562 'run':1563 'busy':1564 'map':1565 'picture':1566 'gaming':1567 'vertical':1568 'playing tennis':1569
'75':1570 'air france':1571 'diesel':1572 'elmo':1573 'crow':1574 'piano':1575 'san diego':1576 'brushing':1577 'colorado':1578 'jumping':1579
'p':1580 'basket':1581 'dark':1582 'hundreds':1583 '36':1584 'flip phone':1585 'heels':1586 'pizza':1587 'speaker':1588 'cut':1589
'oregon':1590 'united':1591 'yellow and white':1592 'kiting':1593 'garbage':1594 'toothbrushes':1595 'barrier':1596 'parking meter':1597 'japanese':1598 'shoes':1599
'construction':1600 'in air':1601 'lily':1602 '2016':1603 'private':1604 'falling':1605 'trunk':1606 'tour':1607 'soap':1608 'noon':1609
'bacon':1610 'on ground':1611 'clean':1612 'catholic':1613 'surf':1614 'flat screen':1615 'wii':1616 'on floor':1617 '3 feet':1618 'flour':1619
'blue and green':1620 'flower':1621 'baseball glove':1622 'paper towels':1623 'on plate':1624 'beets':1625 'on tracks':1626 'zebras':1627 'batman':1628 'ski pole':1629
'finger':1630 'behind':1631 'green beans':1632 'stone':1633 'cups':1634 'wilson':1635 'boots':1636 'hills':1637 'pedestrian':1638 'bell':1639
'hats':1640 'north face':1641 'window':1642 'eat':1643 'logitech':1644 'bottom right':1645 'mixer':1646 'target':1647 'foreground':1648 'in middle':1649
'hit':1650 'fabric':1651 'all of them':1652 '193':1653 'crocs':1654 'volvo':1655 'teddy':1656 'cobblestone':1657 'wings':1658 'sparrow':1659
'syrup':1660 'uniform':1661 'white and black':1662 'shark':1663 'panda':1664 'rocky':1665 'apple':1666 'privacy':1667 'us air force':1668 'traveling':1669
'video game':1670 'country':1671 '6:30':1672 'canoe':1673 '12:40':1674 'biker':1675 'backwards':1676 'shih tzu':1677 '2:35':1678 '48':1679
'farmers market':1680 'daisy':1681 'to see':1682 'red and white':1683 'hit ball':1684 'wheel':1685 'towel':1686 'onion rings':1687 'horns':1688 'throwing':1689
'shingles':1690 'burgers':1691 'british':1692 'lacoste':1693 'pie':1694 'cutting':1695 'canon':1696 'pony':1697 'broke':1698 'tell time':1699
'1:05':1700 'baseball game':1701 'pepperoni':1702 'whole':1703 'wires':1704 'gold':1705 '20 ft':1706 'thailand':1707 'wiimote':1708 'top left':1709
'bear':1710 'footprints':1711 'mug':1712 'feeding':1713 'slacks':1714 'camera':1715 'coach':1716 '1 way':1717 'luggage':1718 'shower':1719
'garden':1720 'sliced':1721 'playing baseball':1722 'train':1723 'lift':1724 'window sill':1725 'utensils':1726 'birthday cake':1727 'pumpkin':1728 'spaghetti':1729
'cirrus':1730 '41':1731 'meat':1732 'sony ericsson':1733 'donkey':1734 'ship':1735 'laying':1736 'i don't know':1737 'cow':1738 'grizzly':1739
'f':1740 'mother':1741 'not at all':1742 'log':1743 'vegetable':1744 'r':1745 'top':1746 'neck':1747 'ski':1748 'bank':1749
'anniversary':1750 'trains':1751 'pen':1752 'baseball cap':1753 'beijing':1754 'ramp':1755 'tablecloth':1756 'pineapple':1757 'railroad crossing':1758 'ski resort':1759
'protest':1760 '12:05':1761 'cell phones':1762 'post':1763 'blue jay':1764 'mickey mouse':1765 'island':1766 '1980':1767 'marble':1768 '19':1769
'harley':1770 'black and brown':1771 'orchid':1772 'in field':1773 'st patrick's day':1774 'pacific':1775 'rainbow':1776 'baby's breath':1777 '43':1778 '9':1779
'turning':1780 'fun':1781 'girls':1782 '38':1783 'room':1784 'dad':1785 'stripes':1786 '1950s':1787 '12:45':1788 'mountain dew':1789
'yard':1790 'rack':1791 'united states':1792 'brazil':1793 'pink and blue':1794 'on skateboard':1795 '10:40':1796 'tiles':1797 'trailer':1798 '2:30':1799
'bushes':1800 'asparagus':1801 'shopping':1802 'evergreen':1803 'phone':1804 'coffee pot':1805 'palm trees':1806 'roast beef':1807 'seafood':1808 'plastic':1809
'teddy bears':1810 'tires':1811 'grilled cheese':1812 'on left':1813 '100':1814 'festival':1815 'napkins':1816 '45':1817 'stuffed animal':1818 'fire extinguisher':1819
'7:45':1820 'san francisco':1821 'fashion':1822 'holding baby':1823 'switzerland':1824 'club':1825 'angels':1826 'recently':1827 '1 on right':1828 'cat and dog':1829
'seagulls':1830 '4:45':1831 'cooked':1832 'jackets':1833 'sausage':1834 'bald':1835 '1:15':1836 'face':1837 'ribbon':1838 'dining':1839
'flag':1840 'letters':1841 'mango':1842 'nowhere':1843 'powdered':1844 'smile':1845 'brown and white':1846 'legs':1847 'hitting':1848 'bib':1849
'tail':1850 'french':1851 'numbers':1852 'strawberries':1853 'seeds':1854 'sauce':1855 'going':1856 '1 in back':1857 'laughing':1858 'dishes':1859
'cooler':1860 '10':1861 '3:30':1862 'maple leaf':1863 'bunny':1864 'best buy':1865 'students':1866 '12:00':1867 'stroller':1868 'on road':1869
'glove':1870 'hot dogs':1871 's':1872 'second':1873 '42':1874 'it's raining':1875 'heart':1876 'coming':1877 'rainy':1878 'paddle boarding':1879
'not high':1880 'dr pepper':1881 'african american':1882 'very big':1883 '12:20':1884 'parakeet':1885 'bad':1886 'ottoman':1887 'italy':1888 'photoshop':1889
'1:00':1890 'display':1891 'catching':1892 'water skis':1893 '2nd':1894 'foot':1895 'pacifier':1896 'tulip':1897 'it isn't':1898 'polar bear':1899
'subway':1900 'on woman':1901 'dock':1902 'choppy':1903 'grazing':1904 'flats':1905 'pizza cutter':1906 'fires':1907 'buttons':1908 'drink':1909
'waving':1910 'marshmallows':1911 'cabbage':1912 'prom':1913 'homemade':1914 'no cat':1915 'bottles':1916 'facebook':1917 'mirrors':1918 'westin':1919
'8:50':1920 'jal':1921 '11:25':1922 '12 feet':1923 'index':1924 'dead':1925 'police officer':1926 '6:40':1927 'power':1928 '1 year':1929
'sideways':1930 'on sign':1931 'driveway':1932 'cone':1933 '1:10':1934 'first':1935 'windows':1936 'magazines':1937 '52':1938 'orange and blue':1939
'on his head':1940 'sailing':1941 'drawer':1942 'dump truck':1943 'daffodil':1944 'colgate':1945 'amtrak':1946 'mitt':1947 'tropical':1948 'golf':1949
'iris':1950 'ireland':1951 'squash':1952 'blue and white':1953 'nighttime':1954 'candy':1955 'white and gray':1956 'heinz':1957 'white and blue':1958 'it's not':1959
'bike rack':1960 'in snow':1961 'refrigerators':1962 'citizen':1963 'raft':1964 '2011':1965 'park':1966 '2 people':1967 'cauliflower':1968 'lemons':1969
'wave':1970 'red and silver':1971 'kayaking':1972 'cap':1973 'gray and red':1974 'skull':1975 'teeth':1976 'balloon':1977 'tarp':1978 'india':1979
'used':1980 'queen':1981 'purple':1982 'hot dog':1983 'pedestrian crossing':1984 'station':1985 'green and black':1986 'himself':1987 'soccer ball':1988 'flickr':1989
'trees':1990 'tiled':1991 'mat':1992 'whipped cream':1993 'labrador':1994 '10:15':1995 'recessed':1996 'yellow and blue':1997 'conductor':1998 'brushing teeth':1999
'donuts':2000 'love':2001 'big':2002 '3:50':2003 'duck':2004 'life':2005 '5 ft':2006 'on dresser':2007 'fisheye':2008 '4:15':2009
'boundaries':2010 'king':2011 'skateboards':2012 'foil':2013 'dogs':2014 'wax':2015 'on right':2016 'in hand':2017 'castle':2018 'taking selfie':2019
'green and white':2020 'slow':2021 'santa hat':2022 'army':2023 'rolex':2024 'onion':2025 'yellow and black':2026 'mario':2027 'egg':2028 '150':2029
'name tag':2030 'life vest':2031 'on beach':2032 'fresh':2033 '2:55':2034 'tissues':2035 'party':2036 'corona':2037 'elm':2038 '7:25':2039
'radiator':2040 'oil':2041 'lid':2042 'yellow and orange':2043 'baseball field':2044 'mouth':2045 '1 hour':2046 'decoration':2047 'cool':2048 'baking':2049
'racing':2050 'skate':2051 '51':2052 'lanes':2053 'writing':2054 'calico':2055 'bride':2056 'bottom left':2057 'on bench':2058 'salt':2059
'english':2060 'blonde':2061 'arch':2062 '33':2063 'above':2064 'goalie':2065 'cameraman':2066 'star':2067 'nobody':2068 'eyes':2069
'playing wii':2070 'hydrant':2071 'cup':2072 'women's':2073 'pooping':2074 'chains':2075 'cupcakes':2076 'ear':2077 '12:28':2078 '11:10':2079
'palm tree':2080 '12:15':2081 'roman':2082 'texas':2083 'snowy':2084 'middle':2085 'nikon':2086 'semi':2087 'ski boots':2088 'feeding giraffe':2089
'silver and black':2090 'pitbull':2091 'john':2092 'several':2093 'living':2094 'playing video games':2095 '10:08':2096 '66':2097 'stuffed':2098 '64':2099
'cheesecake':2100 '8:00':2101 'oak':2102 'raining':2103 'cones':2104 'little girl':2105 'air canada':2106 'tree branch':2107 'above stove':2108 'london':2109
'straight ahead':2110 '4:00':2111 'bow tie':2112 '11:20':2113 'on his face':2114 'tile':2115 'roundabout':2116 'on elephant':2117 'giraffe and zebra':2118 '7:10':2119
'hungry':2120 'women':2121 'pillow':2122 '4:05':2123 'toaster':2124 'shallow':2125 'in back':2126 '1950':2127 'under sink':2128 'empty':2129
'next to toilet':2130 'babies':2131 'holding phone':2132 'thumb':2133 'black and yellow':2134 '11:00':2135 'swing':2136 'american airlines':2137 'low':2138 'under armour':2139
'm':2140 'floating':2141 'looking out window':2142 'mohawk':2143 'ups':2144 'south':2145 'white and yellow':2146 'long sleeve':2147 'very':2148 'bank of america':2149
'on rack':2150 'polka dots':2151 'no man':2152 '3rd':2153 'green and orange':2154 'printer':2155 'tv':2156 'delta':2157 'b':2158 'blue and pink':2159
'closet':2160 'e':2161 'over easy':2162 'in vase':2163 'woman's':2164 '2008':2165 'soon':2166 '8:35':2167 'medium':2168 'smaller':2169
'chest':2170 'top right':2171 '500':2172 'to get to other side':2173 'in his hand':2174 '1 on left':2175 'zebra and giraffe':2176 'pomeranian':2177 'no grass':2178 '2:45':2179
'vests':2180 'tripod':2181 '4:20':2182 'rubber':2183 'windy':2184 '2:00':2185 'take off':2186 '90':2187 '3:10':2188 'hazy':2189
'smooth':2190 'not here':2191 'first base':2192 '9:25':2193 'remotes':2194 'soccer field':2195 'on bus':2196 'miami':2197 'rackets':2198 'for balance':2199
'cord':2200 'at table':2201 'flying kites':2202 'spring':2203 'very tall':2204 'v':2205 'corgi':2206 'asia':2207 '3:20':2208 'in suitcase':2209
'hispanic':2210 'men's':2211 '8:55':2212 'powdered sugar':2213 'behind clouds':2214 '11:45':2215 '2:25':2216 'control':2217 'he's not':2218 'not long':2219
'1:45':2220 'i':2221 'blueberries':2222 '2015':2223 'watch tv':2224 'drinking water':2225 'on wall':2226 'washington':2227 'to right':2228 'japan':2229
'2 years':2230 'hot sauce':2231 'in her hand':2232 'on table':2233 'man on left':2234 'man on right':2235 '4:35':2236 '4:30':2237 '7:05':2238 'very fast':2239
'1 foot':2240 'owner':2241 '2 hours':2242 '5:55':2243 'tennis shoes':2244 'never':2245 'in corner':2246 'boredom':2247 'size':2248 '1990':2249
'cheddar':2250 'australia':2251 '7:00':2252 'street light':2253 'shelves':2254 'no smoking':2255 '4:50':2256 'l':2257 'on bed':2258 'on train':2259
'20 feet':2260 '1:40':2261 '12:25':2262 'pee':2263 'us airways express':2264 '6:25':2265 'in background':2266 '5 feet':2267 'alps':2268 'coaster':2269
'1:20':2270 'happiness':2271 '8:05':2272 '11:50':2273
组会
一张图片对应多个问题,如果仅仅使用蕴含,是不是会相互影响?
不同n对不同类型问题的影响是不同的,不同模型也是不同的?
Check It Again:论文整理相关推荐
- 关系抽取论文整理,核方法、远程监督的重点都在这里
来源 | CSDN 博客 作者 | Matt_sh,编辑 | Carol 来源 | CSDN云计算(ID:CSDNcloud) 本文是个人阅读文章的笔记整理,没有涉及到深度学习在关系抽取中的应用. 笔 ...
- 论文整理集合 -- 吴恩达老师深度学习课程
吴恩达老师深度学习课程中所提到的论文整理集合!这些论文是深度学习的基本知识,阅读这些论文将更深入理解深度学习. 这些论文基本都可以免费下载到,如果无法免费下载,请留言!可以到coursera中看该视频 ...
- Non-Blind图像反卷积论文整理
Non-Blind图像反卷积论文整理 1 Spatial Deconvolution Stochastic Deconvolution 2013 http://www.cs.ubc.ca/lab ...
- 计算机维修知识综述论文,机器学习领域各领域必读经典综述论文整理分享
原标题:机器学习领域各领域必读经典综述论文整理分享 机器学习是一门多领域交叉学科,涉及概率论.统计学.逼近论.凸分析.算法复杂度理论等多门学科.专门研究计算机怎样模拟或实现人类的学习行为,以获取新的知 ...
- 内窥镜去反光的论文整理
文章目录 内窥镜去反光的论文整理 Detection and correction of specular reflections for automatic surgical tool segmen ...
- ECCV2020超分辨率方向论文整理笔记
ECCV2020超分辨率篇 ECCV的全称是European Conference on Computer Vision(欧洲计算机视觉国际会议) ,是计算机视觉三大顶级会议(另外两个是ICCV]和C ...
- 流量分类方法设计(一)——参考论文整理
流量分类方法设计(一)--参考论文整理 因为最近在做流量分类有关的工作,所以将整个工作思路整理下来,希望对以后进一步的学习和论文写作有所帮助. 这一篇主要整理一下最近有关流量分类的论文,介绍他们的设计 ...
- WSDM'23 | 工业界搜推广nlp论文整理
大家好,蘑菇先生. WSDM'23已公布录用结果,共收到投稿690篇,录用123篇,录用率为17.8%,完整录用论文: https://www.wsdm-conference.org/2023/pro ...
- ICCV2019超分辨率方向论文整理笔记
ICCV2019超分辨率篇 IEEE International Conference on Computer Vision,即国际计算机视觉大会,与计算机视觉模式识别会议(CVPR))和欧洲计算机视 ...
最新文章
- Ubuntu 14.04 LTS, 64bit, cuda 7, Caffe环境配置编译和安装
- PE 学习(七)第七章:资源表
- mysql交互式创建表_用mysql语句创建数据表详细教程
- React setStats数组不更新,百思不得其解。
- cluster+raid5+oracle,raid5的一块盘状态为missing
- 若依集成ueditor富文本编辑器
- 下载qq付费音乐的demo
- 【题解】LuoGu4408:[NOI2003]逃学的小孩
- python风变编程和扇贝编程_想学习phython ,纠结是扇贝编程还是风变编程?
- 优化概率神经网络_贝叶斯神经网络最新综述
- 浅谈Clean Code
- Comparing JSON and XML? Stop it!
- 我叫mt4服务器注册 满了,我叫MT4注册上限怎么办 人数上限解决办法
- 青蛙的约会(ojld)
- c语言流水灯程序 16,16个发光二极管流水灯程序
- APP第三方安全检测
- Eclipse护眼背景色图文设置
- 免费开放API支付接口
- 图码详解算法|理解八大思想,胜刷百道力扣!
- 超分辨率数据集的下载
热门文章
- Java程序与以太坊智能合约交互
- FPGA学习日志——一位全加器full_adder
- View androidx.appcompat.widget.AppCompatImageViewdoes not have a NavController set
- ¥9.9拼5节scratch青少儿编程课程,还能领取少儿编程大礼包!快分享吧!
- 2023最新盲盒H5商城系统小程序源码+支持抽奖功能/VUE内核开发
- 浅谈oracle树状结构层级查询测试数据
- 【乐逍遥网站设计】营销型企业网站设计制作6要素
- .ttf字体文件引入vue项目及使用
- 使用OpenStreetMap,Hot export tool,OSMNX下载地图资源
- Unity3D 调用其他脚本函数方法