Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

A higher result than paper, but i don't know why. #4

Open
GuideWsp opened this issue Jul 13, 2021 · 9 comments
Open

A higher result than paper, but i don't know why. #4

GuideWsp opened this issue Jul 13, 2021 · 9 comments

Comments

@GuideWsp
Copy link

I don't know why,but I got a 84.42%@vip_s7,the following is the cmd and log.
Note: I train the vip with 6 Ti1080, batch=64.

CUDA_VISIBLE_DEVICES=0,1,2,4,6,7 ./distributed_train.sh 6 --data /data/path/to/imagenet --model vip_s7 -b 64 -j 6 --opt adamw --epochs 300 --sched cosine --apex-amp --img-size 224 --drop-path 0.1 --lr 2e-3 --weight-decay 0.05 --remode pixel --reprob 0.25 --aa rand-m9-mstd0.5-inc1 --smoothing 0.1 --mixup 0.8 --cutmix 1.0 --warmup-lr 1e-6 --warmup-epochs 20

epoch,train_loss,eval_loss,eval_top1,eval_top5
0,6.897498446451107,6.920823193237253,0.11199104697234573,0.5319574684269399
1,6.635074172221439,6.908653669171921,0.14598833069916906,0.7299416333558549
2,6.299999626589493,6.877433327481934,0.1879849714256746,1.049916047514386
3,5.961820629281058,6.803794402197336,0.42396609998304324,1.7018639201422174
4,5.721933814841257,6.7456617523561375,0.5619550690766516,2.487801056386528
5,5.513945505652629,6.616568194152012,1.1479082055191414,4.049676146414574
6,5.295299509881248,6.326601969986933,1.9978402258825154,6.5834735274629566
7,5.135693771738402,5.957425979278286,3.4817215823600813,10.541156985532588
8,4.9750422289673715,5.509230036514682,6.265498933301202,17.146628777357723
9,4.955138535566733,4.902270209969506,12.071034721885335,28.171747068950303
10,4.849730646106559,4.18796934782815,20.450364618527967,42.05063708428716
11,4.66967850000086,3.578551411771763,29.06567554780554,53.971683791480764
12,4.74806534404486,3.138725295637848,36.47908275797409,62.429007578502606
13,4.681437905405609,2.8611019714461663,41.726663123969125,67.90856960050374
14,4.665561343582583,2.694952508028331,45.61235242660079,71.53227981629634
15,4.644043553043419,2.5987647279492667,48.32413533427946,73.82609615046714
16,4.6149375808071085,2.52884503917955,50.23598260012124,75.46396516855312
17,4.571690149710212,2.461466482797343,51.871851789942745,76.69786689251139
18,4.573833277527715,2.36428716104323,53.73170294325673,78.08775552594464
19,4.600005122977243,2.263626220273777,55.1275915150909,79.41364968878662
20,4.586769611063138,2.160541468110621,56.7774594400551,80.81153736430333
21,4.567850512518009,2.0593713660705912,58.22534356782097,81.88945123861527
22,4.563050458129023,1.9497668935951067,59.70922500577395,82.97336470528647
23,4.49174424292336,1.8379585912881609,61.295098270190145,84.19926647079171
24,4.531819054778193,1.7433091885814114,62.55499736107537,85.25118260009414
25,4.544238956881241,1.6586213802473133,63.780899418610474,86.10711376244197
26,4.501337981559861,1.5895183448571795,64.77482004442574,86.7730603658641
27,4.434042557864122,1.5319257438225742,65.63675117538449,87.31701739983848
28,4.5165048317170475,1.4855170154388442,66.23270318672358,87.74098337208707
29,4.477951758344409,1.4469930333200907,66.78065965294066,88.03095997864604
30,4.414669812565118,1.4166177173927588,67.1686282587017,88.26894115377507
31,4.408424125590795,1.3925092366206895,67.47260409474859,88.53891952803932
32,4.4501715982464,1.3740543007764823,67.75258193944094,88.7189051877803
33,4.436891948673087,1.3572717362924076,68.01856057203901,88.88689186603354
34,4.47835459507687,1.343482419458659,68.24654237377807,89.01488126128132
35,4.398498310169703,1.3305243780380076,68.3825315617817,89.14287107946892
36,4.421788746202496,1.3195938538784084,68.58451535302957,89.26686126950665
37,4.420958277205346,1.3100458558926857,68.7265037498172,89.35885389056571
38,4.417498296415302,1.3024063418173006,68.84249466707702,89.50484254350472
39,4.377080974444537,1.2944942471618872,69.00448140418298,89.6288324094717
40,4.384807224005041,1.2872051226883905,69.17446816657011,89.7348238325016
41,4.350281732183107,1.2810902684204637,69.31245737279494,89.79081968486466
42,4.38020530888732,1.2767441393461105,69.42244829764205,89.86481388129575
43,4.358453169674941,1.271780503977681,69.59243459131667,89.88881151470773
44,4.372297901502797,1.2678588287476187,69.70842538041293,90.01480166987032
45,4.352815063906387,1.2648489257257736,69.81641682898767,90.08479558085548
46,4.3347624590699105,1.2628966046661274,69.8984102414385,90.13679155789607
47,4.372627782150054,1.2602339647987464,69.94640632484753,90.16278951712184
48,4.336176442428374,1.2586284394970264,70.08239564101473,90.2187850362596
49,4.3249931939890685,1.256730695808709,70.1683887759072,90.28677945998695
50,4.387032535714163,1.2546128732339352,70.29237887623043,90.33077638945134
51,4.292173899395365,1.252888638251744,70.30837772846337,90.36677337419452
52,4.334522573041244,1.25077576496517,70.4223681404804,90.40077038099648
53,4.343562552626704,1.2486651011843002,70.46036552061187,90.44876675735082
54,4.2876138149852485,1.2462088407187373,70.55435769209508,90.48276395822904
55,4.33589161953456,1.2429493092105977,70.65035050613919,90.53475971922246
56,4.266452463579849,1.2397431261338632,70.74834232957409,90.55075857603266
57,4.326588875810865,1.2369515137733074,70.7603412076155,90.5827559381187
58,4.357701543351294,1.233855538497533,70.79433852292549,90.64675089664702
59,4.318143364409326,1.2319323991968216,70.93032786380996,90.68474789594957
60,4.314157845268787,1.2295630219058793,71.00432144505588,90.72874470823584
61,4.297892332077026,1.2265230200689474,71.10431390464349,90.81073740663247
62,4.281018173190909,1.224780632947771,71.13831120439076,90.83273619360452
63,4.24098925523355,1.2214524747657638,71.30029815754388,90.85073457795251
64,4.301690471004433,1.2191791983568498,71.33829531430455,90.87873241625044
65,4.255715172055742,1.2171746795343958,71.38829123284395,90.89073145632725
66,4.281312667148214,1.2140680893107745,71.46028553189377,90.91072970143834
67,4.293928156436329,1.2116449428514409,71.60027411918375,90.98672405157782
68,4.277459722169688,1.2099077029854153,71.65027039291935,91.02072099979073
69,4.311910683000591,1.2092339148588938,71.73626323028924,91.06071793675451
70,4.237875575750646,1.2071373786232813,71.79225863224887,91.06271781582669
71,4.29129040073341,1.2068184426799506,71.79425837583383,91.08471543101747
72,4.292811004208847,1.2045458831002107,71.86825275901757,91.17670852078903
73,4.2103998963262,1.202459813423056,71.91024979842433,91.21070546900194
74,4.268061406175855,1.2004286657857013,72.01824037548676,91.2327035712143
75,4.276580367289799,1.1972193041092662,72.03023983667244,91.25670206149132
76,4.1927781541582565,1.194762274204526,72.12423260045446,91.30069840506509
77,4.223329272068722,1.192597238400127,72.1082336850552,91.32469660239681
78,4.239088642765099,1.1889550637821227,72.18622731189805,91.35469416230882
79,4.239531100635797,1.1856474060505335,72.25822169700056,91.38669133031857
80,4.253520881625968,1.182930325586389,72.36021350835878,91.41668928204494
81,4.264392399452102,1.180294969627146,72.42020867944822,91.46668532142428
82,4.2935264076985105,1.1780071623582247,72.4242081336618,91.54467927599971
83,4.183295256654981,1.1759697122770676,72.55419775402343,91.54667892071564
84,4.1724416101482555,1.173063437430041,72.67818824616101,91.58467580284007
85,4.253374935875477,1.1703134040223742,72.76418096635277,91.63467209488475
86,4.173966360763765,1.1681256303691108,72.82817604205921,91.6386715600838
87,4.203923020564335,1.165260956960815,72.89217057947864,91.67066915652607
88,4.198879456855882,1.1630617950584552,72.9921627168264,91.68866748228501
89,4.204212742792049,1.1595675451795002,73.05015833108351,91.76666124278417
90,4.220558797809439,1.1575091804049413,73.14615044663614,91.78265986523813
91,4.234742809349383,1.1546186548301234,73.23614321791757,91.81865680970132
92,4.211457870375942,1.151942806447585,73.24414290997345,91.7766602280216
93,4.176632179340846,1.1492926530595418,73.28013967866946,91.8186568499813
94,4.13435952428361,1.1468784104785046,73.32213649470542,91.85265428521578
95,4.150933070921562,1.1448545083457227,73.37213264027746,91.85665412025827
96,4.212816624574258,1.1424470642415745,73.41812883545977,91.89265110500145
97,4.1783625945239,1.1403365648991488,73.47612398100438,91.89665058850956
98,4.149515078101359,1.1381833895179103,73.51612139766611,91.95864578514383
99,4.184193624577052,1.135834064362345,73.52012092145421,91.99664300049355
100,4.1438532412891655,1.1334068423871488,73.57611648087196,92.0086419417013
101,4.2061415289489315,1.131613763940114,73.59811418394634,92.05263816809695
102,4.1770783679586065,1.128588039811521,73.66810895545837,92.07463654494555
103,4.130896655606552,1.1262294837030713,73.7041060628724,92.14463090633413
104,4.164088279428617,1.1236643548031233,73.79009931036572,92.19462712148068
105,4.1543260863129525,1.1212902866967267,73.89409091291252,92.23062394876574
106,4.121105573546719,1.118689010662228,73.96008545756769,92.2386235236435
107,4.152086059812089,1.1163591871680416,74.07007676690553,92.25462214609745
108,4.112779476273228,1.1132477346529686,74.08007607621371,92.25062268089842
109,4.138080627145902,1.1113340086521564,74.17006893537872,92.30661806454897
110,4.109547010609801,1.1093562793067986,74.19806683226572,92.35461428344522
111,4.1400179493595175,1.1067077219271677,74.17206843362199,92.43460804019468
112,4.122481537536836,1.1039926298293445,74.2740603328638,92.42660852390598
113,4.130142235420119,1.1019119612455996,74.35605391009638,92.4746044498569
114,4.106339105418031,1.0986333671793653,74.43204776222882,92.53659964649117
115,4.10345072813437,1.0962525172172932,74.53603968518456,92.54259916652957
116,4.145501025965516,1.0950553909262393,74.63003195095955,92.56259748853881
117,4.055538654327393,1.092459419948293,74.63803127317198,92.5325999066559
118,4.107625719527124,1.0902880965476132,74.69802623919969,92.55659773414416
119,4.097143169859765,1.0863236726431758,74.75002239200747,92.68258761467051
120,4.103760034265653,1.0848064310464294,74.78002005078851,92.7085856910744
121,4.094524071250163,1.0829915166444735,74.83401578972315,92.68458761092081
122,4.0647215876780765,1.0807031207671691,74.87601228351927,92.76458164230651
123,4.111846302596616,1.0778105026073013,74.94800669974722,92.8305760111945
124,4.051295757293701,1.075635140280315,75.0020022043256,92.87657223384043
125,4.074876647599986,1.073448331057954,75.04999838294187,92.89057127016753
126,4.065043056514901,1.0720190838254668,75.08399538974382,92.92056875318139
127,4.084583755949853,1.0708788350700102,75.1779882999985,92.92856826947008
128,4.036718828577391,1.0690497463488138,75.17998821935066,93.00056245134178
129,3.981601013264186,1.066288032891054,75.28198010760701,92.9965627517865
130,4.079461373073954,1.0648121422133001,75.35997348727727,92.99456281412526
131,4.064721067186812,1.0631430921492773,75.3959705965222,93.04255920878869
132,4.068798142419735,1.0618841971943698,75.44196721830613,93.05255825261519
133,4.034127584645446,1.0586237160170253,75.53795944005144,93.13055220719062
134,3.9986095159826145,1.056434438444311,75.58195622304318,93.15255005673765
135,4.0494951160860735,1.0536063259844304,75.6919472870393,93.19654663466767
136,4.010537627717139,1.0510139629541613,75.77594024716792,93.2345438280465
137,3.952524077724403,1.0484931318771933,75.75594183544418,93.2405430148596
138,3.992410169520848,1.0460639876486273,75.90992977214655,93.26054180558134
139,3.979717647525626,1.0429913537348379,75.91992850288774,93.31453719298159
140,4.027233473012145,1.0405188840672013,75.97392467574763,93.32053665443092
141,4.0813811396209285,1.0375410683698039,76.02792021511326,93.36253317752158
142,3.9673641164537887,1.034782726930033,76.09991454528853,93.37853226868803
143,3.9360495486729583,1.0329528738618727,76.12391277008386,93.42452866710116
144,3.982970573532749,1.031459635815118,76.2259044075058,93.424528630483
145,4.010461907991221,1.0283353227432723,76.29989843915514,93.5025224678803
146,3.9713859087984327,1.0256409141866887,76.33789559042307,93.49452330312599
147,3.9882485933706793,1.0228629822864903,76.37789218683794,93.5425190716188
148,3.9469488439425615,1.0196745478623428,76.46388505350237,93.57651641164605
149,3.972708628211223,1.0163749271312261,76.5638773849264,93.63051203340257
150,3.9860113741646352,1.012866180578753,76.58387576735561,93.69050676507403
151,3.9316770627465045,1.009717572527822,76.68186806591117,93.74450299103025
152,3.995447256195713,1.006326335381054,76.79585906381884,93.76850124695102
153,3.9208318649883003,1.0038157002028156,76.81585723935542,93.82449610330002
154,3.966935953623812,1.0022543991730601,76.8718528280677,93.82249634140597
155,3.900635373424476,1.000155895025804,77.0498383633948,93.89249083828176
156,3.9117869793529243,0.998312960948079,77.1458308304818,93.93648741621178
157,3.9204770645625153,0.9957889344499413,77.173828503998,93.97248465362028
158,3.9169989001583048,0.9928011367267804,77.20982563155202,94.00848159808348
159,3.9202801711122754,0.9887251343863286,77.2998185493061,94.0104815540538
160,3.906469684251597,0.9861672291917025,77.37181281906139,94.03447967448737
161,3.9593827623716544,0.983605310446014,77.42980836374399,94.06847699254372
162,3.876635793229224,0.9816954550709918,77.46580536679625,94.10447421164315
163,3.9001552756403535,0.9796518905288114,77.49780249452603,94.14847049662785
164,3.835129989704616,0.9770739230143815,77.57379640524753,94.1584697748106
165,3.898776393541148,0.9746265718626238,77.61979292083879,94.1744686316208
166,3.916073674886999,0.9713456796346764,77.66978896021814,94.23046389809323
167,3.854554978894516,0.968392954846495,77.72378455268012,94.29245927049469
168,3.8261295976773115,0.9656309809420416,77.78577933919095,94.32845609777975
169,3.8342656820592746,0.9632433774175477,77.92376818289596,94.40045051950044
170,3.8310181825933323,0.9598682177905016,78.00576187730665,94.46644512274467
171,3.8278482430417773,0.9563763968401571,78.08975523474236,94.50044233643926
172,3.8315204096512057,0.9536149256344405,78.19574680424492,94.51444107982104
173,3.815767133739633,0.9502094194761477,78.2917392420374,94.55643795444607
174,3.8109556050367757,0.9475126778561633,78.31773717196863,94.60443399757511
175,3.798724056969226,0.944900313612652,78.37973222213024,94.61043386914788
176,3.8188754739895674,0.9417507125228779,78.465725520889,94.61243374822007
177,3.7906133960670148,0.9377580324694325,78.4777239640897,94.61243348273837
178,3.841992200260431,0.9353244338700432,78.59571483731985,94.62443275717142
179,3.8163234885309785,0.9324265027864391,78.68970713238937,94.68842724698723
180,3.8143192613628547,0.9294543408191773,78.73970324134324,94.73642334870534
181,3.7683921229671427,0.9261675824185255,78.8036984342278,94.78241980570753
182,3.7818730481913394,0.923982460521124,78.83169595028589,94.84441482657462
183,3.755684365688915,0.9212717203776957,78.88969158285208,94.85041461209471
184,3.811672761406697,0.9191701354698967,78.96368532155611,94.87241293035424
185,3.7187630223556303,0.9165008361030031,79.07167678843993,94.91240963296178
186,3.7433232388026276,0.9128048724913043,79.16366926302634,94.94440711222595
187,3.783568885964407,0.9098437855447944,79.193666969411,94.97040483709546
188,3.7836115662480743,0.9066717340873534,79.25566220632525,94.9604057346799
189,3.7132954631053225,0.9037563167718037,79.32965616840008,94.99040315910469
190,3.7004594366315384,0.9007076963349615,79.47364497156146,95.0224008727251
191,3.7304701771534665,0.8978327910105387,79.55763772479744,95.08239607310907
192,3.700018348828168,0.8942069873439056,79.58763541287303,95.15439033187891
193,3.6864709585485325,0.8912874848430095,79.66762916962249,95.17638853296033
194,3.716994675112442,0.8881717415373188,79.77762017502955,95.21238559460166
195,3.6630856252052415,0.8846277142989999,79.83161617761498,95.24038335600144
196,3.70445132591355,0.8816261793859348,79.90161046942904,95.2903792196136
197,3.667574173967603,0.8784933531916568,79.9416069962694,95.34837442374727
198,3.673860294718138,0.8758927606294731,80.07559646265896,95.36637298386246
199,3.647094535156035,0.8726708035918772,80.16358933655913,95.41636914041993
200,3.6449320484215106,0.8685720142827263,80.24158311536738,95.41236961663184
201,3.6660440975511577,0.8651743567416157,80.33757523092001,95.43236791667019
202,3.6045285580863413,0.8614645302081125,80.4675654142859,95.4943630547154
203,3.6495057394806767,0.8582117335241934,80.51956098120304,95.52636053397954
204,3.6250970833738085,0.854794814142072,80.60955384036806,95.57435679315577
205,3.6242427893087896,0.8516537740072072,80.68154798195978,95.60035451802527
206,3.6249839084249147,0.8486535416422972,80.76154147505846,95.64635107389651
207,3.6208607680361036,0.8452714143509998,80.81553724328762,95.69034761154654
208,3.621414886394017,0.8423744141697912,80.89553079497537,95.71434561480199
209,3.595421079178931,0.8396461595297718,80.90353028196954,95.77434071631691
210,3.600605377009217,0.8370815479186103,81.0335199133166,95.81833705989068
211,3.6281545095040766,0.83375701421967,81.12551273943738,95.85633405919323
212,3.604720505190567,0.8295639896667458,81.24350308536596,95.89833075805107
213,3.5827912982081025,0.8255833837377787,81.4014905623346,95.91232946115288
214,3.567787086459952,0.8213720157159613,81.44348708542526,95.95632580472663
215,3.5835306644439697,0.817864267208721,81.5234805831012,96.00632172692787
216,3.5531847208318577,0.8139958393382087,81.59347526215237,96.02032093902214
217,3.556191263064532,0.8101979788756639,81.68346796111291,96.10231420500031
218,3.5177630505091706,0.8064951221806156,81.75546192693744,96.16030967461568
219,3.505100737155323,0.8028557199295516,81.83345549336033,96.17230883736332
220,3.548047834718731,0.7992166765078899,81.90345040035957,96.20830599787368
221,3.5126069230093084,0.7949462399827166,82.04543895002804,96.22230474125547
222,3.4366250877649014,0.791103566382353,82.14943049398578,96.26030164168898
223,3.5024204757851614,0.7875747270998349,82.24342289890978,96.31829690441171
224,3.52920873735992,0.7836198607651427,82.2774204330133,96.35229432133711
225,3.5523327135703933,0.7802904253530079,82.40141039784932,96.3882917345128
226,3.514156509453142,0.7771999123058131,82.44140739340219,96.42828861288753
227,3.480050802230835,0.7738232632063263,82.5373997726056,96.46428532299447
228,3.490331743804502,0.769954922446994,82.67138911083158,96.4842838207709
229,3.4522565519306023,0.7658971668519875,82.7413832342021,96.54027943877772
230,3.4249578193879464,0.7622650279282437,82.76738133532325,96.57427603545622
231,3.437175576115998,0.7588821934200403,82.83737553925373,96.60227404952136
232,3.4281330276542987,0.7550330743933857,82.89737097399396,96.63827128692986
233,3.412930045329349,0.7511200292064099,82.97936447432838,96.69026726397044
234,3.42685407316181,0.7473738250727845,83.14735111230164,96.70226630404724
235,3.4242665700509516,0.7433231376417255,83.16534943806057,96.74026295181542
236,3.406816643728337,0.7396762944652446,83.29333929652816,96.76226162160934
237,3.379652134129699,0.7358025643989741,83.37533269799354,96.80825790284433
238,3.421702680453448,0.7321584496476365,83.42132862769417,96.87225303713984
239,3.4100404222246627,0.7282632348175269,83.57931626212091,96.92824830361228
240,3.3646818852760423,0.7243552193070648,83.67330864507402,96.9402472851
241,3.376362484945378,0.7204217895032234,83.767300910849,96.99224302778434
242,3.3655178748386008,0.7166998286827423,83.88929142504537,97.04423923918118
243,3.366966569927377,0.7125734683561589,84.01128154742737,97.06023780304606
244,3.359389214448526,0.7084448325539596,84.1212729226779,97.10023482972436
245,3.358978704667427,0.7042798031884073,84.22926440054718,97.12023303455548
246,3.30417195172377,0.6996296609595207,84.33325582732678,97.1522305724087
247,3.296816946754993,0.6952631592550294,84.42524891709834,97.20822616295189

@houqb
Copy link
Owner

houqb commented Jul 13, 2021

Hi, thanks for sharing your log with us. Did you train any other models with your dataset and are they normal?

@GuideWsp
Copy link
Author

due to limited gpu, I just try the vip_s7, until to now, the accuracy of vip_s7 is 84.68%。

image

@houqb
Copy link
Owner

houqb commented Jul 13, 2021

Have you ckeched that there is no problem with your val set?

@GuideWsp
Copy link
Author

Yes, I had validated the two model(one is your pretrained, the other is my retrained), the accuracy of the pretrained is 81.472(same as the paper) , while my retrained is 83.922(This is already better than vip_l7)。

CUDA_VISIBLE_DEVICES=3 python3 validate.py --model vip_s7 --checkpoint weights/vip_s7.pth --no-test-pool --amp --img-size 224 -b 64
image

CUDA_VISIBLE_DEVICES=5 python3 validate.py --model vip_s7 --checkpoint output/train/20210712-083848-vip_s7-224/model_best.pth.tar --no-test-pool --amp --img-size 224 -b 64
image

@zihangJiang
Copy link
Collaborator

We are trying to reproduce the issue. The main difference is that you are using a very small batch size (384) while we are using a large batch size (2048).
You can add --use-ema flag to validate your retrained model since ema is used by default for training ViP. I think this will give the 84+% result as shown in your training log. You can also evaluate with imagenetv2-matched-frequency using

python3 validate.py /path/to/imagenetv2 --model vip_s7 --checkpoint /path/to/checkpoint --no-test-pool --amp --img-size 224 -b 64

to see if the Top-1 acc is around 74% (70.9% for our pretrained vip_s7). If so, this is likely to be a much better training receipt for ViP.

@GuideWsp
Copy link
Author

up to now, i have no imagenetv2 datasets. If possible, you can add my Wechat: AIWalker-Happy
The accuracy of the latest retrained model is 85.16(use-ema).
image

@ggjy
Copy link

ggjy commented Jul 22, 2021

Hi, Is there any new progress on this result? The ViP_s7 got the 85.16 ImageNet top-1 or Real top-1?

@iamhankai
Copy link

Any news? Big news, or big bug?

@houqb
Copy link
Owner

houqb commented Aug 1, 2021

Seems that there are some problems with the training data

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants