lombardata/DinoVdeau-large-2024_05_31-batch-size32_epochs150_freeze

DinoVd'eau is a fine-tuned version of facebook/dinov2-large. It achieves the following results on the test set:

• Loss: 0.1235
• F1 Micro: 0.8217
• F1 Macro: 0.7173
• Roc Auc: 0.8829
• Accuracy: 0.3183

---

Model description

DinoVd'eau is a model built on top of dinov2 model for underwater multilabel image classification.The classification head is a combination of linear, ReLU, batch normalization, and dropout layers.

The source code for training the model can be found in this Git repository.

• Developed by: lombardata, credits to César Leblanc and Victor Illien

---

Intended uses & limitations

You can use the raw model for classify diverse marine species, encompassing coral morphotypes classes taken from the Global Coral Reef Monitoring Network (GCRMN), habitats classes and seagrass species.

---

Training and evaluation data

Details on the number of images for each class are given in the following table:

Class	train	val	test	Total
Acropore_branched	1488	465	455	2408
Acropore_digitised	566	169	153	888
Acropore_sub_massive	147	48	48	243
Acropore_tabular	997	290	302	1589
Algae_assembly	2537	859	842	4238
Algae_drawn_up	368	121	131	620
Algae_limestone	1651	559	562	2772
Algae_sodding	3155	980	982	5117
Atra/Leucospilota	1090	359	343	1792
Bleached_coral	219	69	72	360
Blurred	190	63	67	320
Dead_coral	1981	644	639	3264
Fish	2029	657	635	3321
Homo_sapiens	160	63	59	282
Human_object	156	61	53	270
Living_coral	854	289	271	1414
Millepore	383	129	125	637
No_acropore_encrusting	420	153	152	725
No_acropore_foliaceous	204	44	38	286
No_acropore_massive	1017	345	343	1705
No_acropore_solitary	195	54	54	303
No_acropore_sub_massive	1383	445	428	2256
Rock	4469	1499	1489	7457
Rubble	3089	1011	1023	5123
Sand	5840	1949	1930	9719
Sea_cucumber	1413	445	436	2294
Sea_urchins	327	107	111	545
Sponge	269	104	97	470
Syringodium_isoetifolium	1214	388	393	1995
Thalassodendron_ciliatum	781	262	260	1303
Useless	579	193	193	965

---

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• Number of Epochs: 150
• Learning Rate: 0.001
• Train Batch Size: 32
• Eval Batch Size: 32
• Optimizer: Adam
• LR Scheduler Type: ReduceLROnPlateau with a patience of 5 epochs and a factor of 0.1
• Freeze Encoder: Yes
• Data Augmentation: Yes

Data Augmentation

Data were augmented using the following transformations :

Train Transforms

• PreProcess: No additional parameters
• Resize: probability=1.00
• RandomHorizontalFlip: probability=0.25
• RandomVerticalFlip: probability=0.25
• ColorJiggle: probability=0.25
• RandomPerspective: probability=0.25
• Normalize: probability=1.00

Val Transforms

• PreProcess: No additional parameters
• Resize: probability=1.00
• Normalize: probability=1.00

Training results

Epoch	Validation Loss	Accuracy	F1 Macro	F1 Micro	Learning Rate
1	0.17266152799129486	0.23719491234101064	0.7471888818698673	0.51050515939363	0.001
2	0.15843337774276733	0.2506015812994156	0.759063829787234	0.5754792679279648	0.001
3	0.15126323699951172	0.24682021313166036	0.7712364371467471	0.5997966166279454	0.001
4	0.14971894025802612	0.2516328635269852	0.7649153278332611	0.6102972679327003	0.001
5	0.14786843955516815	0.26194568580268135	0.7743272938347361	0.6048101972148001	0.001
6	0.1464792788028717	0.23856995531110348	0.7744215397559949	0.6142379178757014	0.001
7	0.1471046507358551	0.25232038501203163	0.7745991019884542	0.6378919485445709	0.001
8	0.14884118735790253	0.2492265383293228	0.7730045646516785	0.6349994110674028	0.001
9	0.1696147322654724	0.17084908903403231	0.7571320373643019	0.5913962348409779	0.001
10	0.14972557127475739	0.2633207287727741	0.7752799457074991	0.6155340436417148	0.001
11	0.14687682688236237	0.2591955998624957	0.7746945972041475	0.6157336320155493	0.001
12	0.14556235074996948	0.25575799243726366	0.7830250450155508	0.6488857718028787	0.001
13	0.1428324580192566	0.2554142316947405	0.7843104596935376	0.6580664808969154	0.001
14	0.14352969825267792	0.2506015812994156	0.7831899072604227	0.6418882151192971	0.001
15	0.14535032212734222	0.25335166723960123	0.7810577597952191	0.6573171730380419	0.001
16	0.1448182314634323	0.26400825025782054	0.7865547601415288	0.6503641322273297	0.001
17	0.14306068420410156	0.2595393606050189	0.7819386012413047	0.6441131277200526	0.001
18	0.1446852684020996	0.25472671020969406	0.7838684089675614	0.6412796509721743	0.001
19	0.1414887011051178	0.26469577174286696	0.7836073910832345	0.6441719729073186	0.001
20	0.1405678391456604	0.2609144035751117	0.7838809251976828	0.6436752083994657	0.001
21	0.14136537909507751	0.25747679614987967	0.7847450484618627	0.6461144436259285	0.001
22	0.1386082023382187	0.26194568580268135	0.78838833814364	0.6587430664519773	0.001
23	0.13822348415851593	0.268133379168099	0.7900436534586972	0.6502835561459052	0.001
24	0.13893215358257294	0.2650395324853902	0.7880997276346112	0.6519157870757598	0.001
25	0.1396929770708084	0.2695084221381918	0.7883266848624816	0.6553890148296274	0.001
26	0.14012907445430756	0.2650395324853902	0.7793733121865489	0.6349973216617472	0.001
27	0.1402139514684677	0.2499140598143692	0.7915316128502852	0.6667665130743651	0.001
28	0.1389472633600235	0.26400825025782054	0.7907676869041647	0.6575889205208383	0.001
29	0.14010308682918549	0.268133379168099	0.793982620101656	0.6657248014277785	0.001
30	0.13457615673542023	0.2739773117909935	0.7986502613061192	0.6710788331447379	0.0001
31	0.1323619782924652	0.2784462014437951	0.7977876476996564	0.6773801465956272	0.0001
32	0.13328427076339722	0.27294602956342384	0.8018979833926453	0.6778747090268944	0.0001
33	0.13171622157096863	0.2794774836713647	0.8033507506013103	0.6849041252254977	0.0001
34	0.1307307630777359	0.27982124441388795	0.8020612558700079	0.6836107081350358	0.0001
35	0.13025963306427002	0.2853214162942592	0.8046306144154465	0.6903143910655651	0.0001
36	0.1296597272157669	0.290134066689584	0.8050928824879983	0.6931792274102746	0.0001
37	0.12961770594120026	0.288415262976968	0.8034535718733136	0.6884731439645044	0.0001
38	0.12927678227424622	0.28704022000687524	0.8042539049518111	0.68878596648098	0.0001
39	0.13043531775474548	0.2877277414919216	0.8032656478961692	0.6869571453655325	0.0001
40	0.12887024879455566	0.29082158817463044	0.8063839414256921	0.693512791025367	0.0001
41	0.12975196540355682	0.28704022000687524	0.8037148594377511	0.6924207618698955	0.0001
42	0.12854912877082825	0.29597799931247853	0.8086140163056905	0.6899593151302184	0.0001
43	0.12848526239395142	0.28704022000687524	0.8066790352504639	0.6897522012507461	0.0001
44	0.12860073149204254	0.2921966311447233	0.8077876984126985	0.6930773017154849	0.0001
45	0.12759028375148773	0.2956342385699553	0.8106891471599279	0.7036740023156066	0.0001
46	0.12775476276874542	0.2956342385699553	0.8078483318155477	0.6950811210030032	0.0001
47	0.12771955132484436	0.2918528704022001	0.811402081977879	0.7062619936159387	0.0001
48	0.12763886153697968	0.29082158817463044	0.8052384150436536	0.6955184040970307	0.0001
49	0.12784114480018616	0.288415262976968	0.8098617549329287	0.7020626151896329	0.0001
50	0.12698890268802643	0.2939154348573393	0.8101997029212742	0.705975464457343	0.0001
51	0.12740205228328705	0.29769680302509455	0.8129121550109908	0.709798905916108	0.0001
52	0.12743453681468964	0.2918528704022001	0.8121578560339897	0.707621451538995	0.0001
53	0.12782631814479828	0.29322791337229287	0.8083623693379792	0.694158242732084	0.0001
54	0.12641482055187225	0.2939154348573393	0.8095710389288371	0.6945407396863954	0.0001
55	0.12613853812217712	0.29838432451014096	0.8117057825241112	0.711364278725223	0.0001
56	0.12689372897148132	0.2990718459951873	0.8113456464379947	0.7063129571309116	0.0001
57	0.12595032155513763	0.2990718459951873	0.8135370461639561	0.7092631560861781	0.0001
58	0.12691068649291992	0.29254039188724645	0.8129296235679215	0.7038026958133716	0.0001
59	0.1266162097454071	0.2921966311447233	0.8112015199702616	0.7074106748802989	0.0001
60	0.1263982653617859	0.29322791337229287	0.8090558527179996	0.6985455534303677	0.0001
61	0.12581084668636322	0.29597799931247853	0.8097471110366773	0.7041363825305129	0.0001
62	0.12699832022190094	0.2911653489171537	0.8132532581607222	0.7066021008547532	0.0001
63	0.12574061751365662	0.29700928154004813	0.8111204013377926	0.7059955812986827	0.0001
64	0.1252606362104416	0.2939154348573393	0.8139229062217472	0.7090348173219025	0.0001
65	0.12513236701488495	0.2963217600550017	0.8126994653292992	0.70468515245698	0.0001
66	0.125584214925766	0.2949467170849089	0.8140287622403409	0.7118614262032313	0.0001
67	0.12539814412593842	0.2939154348573393	0.8111171298804116	0.7062384835160513	0.0001
68	0.12564098834991455	0.29597799931247853	0.8151627792982313	0.7112630392285837	0.0001
69	0.12584172189235687	0.29941560673771056	0.8142985980159057	0.7140636849121678	0.0001
70	0.12478043138980865	0.29838432451014096	0.8161761696205642	0.718410533722422	0.0001
71	0.12550216913223267	0.2963217600550017	0.8134349886668041	0.7097253169772126	0.0001
72	0.12681567668914795	0.2980405637676177	0.8132986082851795	0.7066535688303756	0.0001
73	0.12942491471767426	0.30044688896528016	0.8154191311441974	0.720052479680047	0.0001
74	0.1252431720495224	0.29597799931247853	0.8141104799538981	0.7156506533830808	0.0001
75	0.12525109946727753	0.29975936748023374	0.8164923076923079	0.7162358218470065	0.0001
76	0.125084787607193	0.2980405637676177	0.8126263668248401	0.7050812600051921	0.0001
77	0.12381099909543991	0.301821931935373	0.8169637369391518	0.7198797063603358	1e-05
78	0.12425024807453156	0.29975936748023374	0.8142230317079229	0.7125350103590883	1e-05
79	0.12345358729362488	0.3028532141629426	0.8181743958197256	0.721102526527663	1e-05
80	0.12395191192626953	0.30216569267789617	0.8199023445381542	0.7272188807325698	1e-05
81	0.12421117722988129	0.3011344104503266	0.816331575477917	0.7213714212722541	1e-05
82	0.1238287091255188	0.3042282571330354	0.8184078588024294	0.7220641511602056	1e-05
83	0.1235181912779808	0.301821931935373	0.8180256808702052	0.7213144754619842	1e-05
84	0.12367285788059235	0.3025094534204194	0.8176539851394697	0.7232957874603048	1e-05
85	0.12321745604276657	0.3045720178755586	0.8164039937288555	0.7142044524891776	1e-05
86	0.1237027570605278	0.303540735647989	0.8165334212478365	0.724290897047381	1e-05
87	0.12352145463228226	0.30216569267789617	0.8180700172173485	0.7200937974757147	1e-05
88	0.12393338233232498	0.303540735647989	0.8187826933214387	0.7208874018784168	1e-05
89	0.12368057668209076	0.3028532141629426	0.8175708900180297	0.717195490689832	1e-05
90	0.12319833785295486	0.30594706084565143	0.8169988469774335	0.7202869112854988	1e-05
91	0.12350083887577057	0.303540735647989	0.8199162022535897	0.7243798285402773	1e-05
92	0.12410824745893478	0.30560330010312825	0.820052770448549	0.7264331892084873	1e-05
93	0.12333476543426514	0.3028532141629426	0.8191851972082453	0.7265708763385225	1e-05
94	0.12396726757287979	0.29769680302509455	0.8147507922788823	0.7139339929197925	1e-05
95	0.12366786599159241	0.2980405637676177	0.8202541859995964	0.7272882506355968	1e-05
96	0.12370884418487549	0.30491577861808183	0.8192505510653931	0.7202747551284554	1e-05
97	0.12347108125686646	0.3007906497078034	0.8185695138296577	0.7212605213289621	1.0000000000000002e-06
98	0.12309076637029648	0.3038844963905122	0.8167741405511973	0.7173148781861652	1.0000000000000002e-06
99	0.12337860465049744	0.30216569267789617	0.8195458231954581	0.7265958923493796	1.0000000000000002e-06
100	0.12339676916599274	0.3025094534204194	0.8176836250613447	0.721153078194628	1.0000000000000002e-06
101	0.12300820648670197	0.3007906497078034	0.8195522327414572	0.727514273407078	1.0000000000000002e-06
102	0.12340469658374786	0.30147817119284975	0.8154413898909484	0.7137375786593034	1.0000000000000002e-06
103	0.12386961281299591	0.29941560673771056	0.8163972286374134	0.7145427089697738	1.0000000000000002e-06
104	0.1235337182879448	0.30147817119284975	0.8170896715732502	0.7188862048208717	1.0000000000000002e-06
105	0.12342803180217743	0.301821931935373	0.8191040415516962	0.7225140266508299	1.0000000000000002e-06
106	0.12395947426557541	0.30216569267789617	0.8179164977705716	0.7182648261775325	1.0000000000000002e-06
107	0.1234135553240776	0.30147817119284975	0.8188311688311688	0.7269935921863244	1.0000000000000002e-06
108	0.12344102561473846	0.29975936748023374	0.8162340337865678	0.722601561126108	1.0000000000000002e-07
109	0.12361280620098114	0.303540735647989	0.8215820979470492	0.7257459935415149	1.0000000000000002e-07
110	0.12370219826698303	0.3031969749054658	0.8204050284975141	0.7232171570839454	1.0000000000000002e-07
111	0.12323758751153946	0.305259539360605	0.8178334500803495	0.7223581742811798	1.0000000000000002e-07

---

CO2 Emissions

The estimated CO2 emissions for training this model are documented below:

• Emissions: 1.717335003598466 grams of CO2
• Source: Code Carbon
• Training Type: fine-tuning
• Geographical Location: Brest, France
• Hardware Used: NVIDIA Tesla V100 PCIe 32 Go

---

Framework Versions

• Transformers: 4.41.1
• Pytorch: 2.3.0+cu121
• Datasets: 2.19.1
• Tokenizers: 0.19.1