lombardata/DinoVdeau-large-2024_07_12-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: 0.1211
• F1 Micro: 0.8216
• F1 Macro: 0.7251
• Roc Auc: 0.8809
• Accuracy: 0.3068

---

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	1472	477	459	2408
Acropore_digitised	565	161	162	888
Acropore_sub_massive	150	47	46	243
Acropore_tabular	997	297	295	1589
Algae_assembly	2545	836	857	4238
Algae_drawn_up	366	129	125	620
Algae_limestone	1650	563	559	2772
Algae_sodding	3147	991	979	5117
Atra/Leucospilota	1084	345	363	1792
Bleached_coral	219	71	70	360
Blurred	190	62	68	320
Dead_coral	1979	642	643	3264
Fish	2022	658	641	3321
Homo_sapiens	161	59	62	282
Human_object	156	56	58	270
Living_coral	400	148	153	701
Millepore	386	128	123	637
No_acropore_encrusting	440	142	143	725
No_acropore_foliaceous	203	39	44	286
No_acropore_massive	1030	344	331	1705
No_acropore_solitary	202	55	46	303
No_acropore_sub_massive	1402	431	423	2256
Rock	4483	1489	1485	7457
Rubble	3093	1006	1024	5123
Sand	5839	1942	1938	9719
Sea_cucumber	1411	433	450	2294
Sea_urchins	328	108	109	545
Sponge	270	104	96	470
Syringodium_isoetifolium	1213	393	389	1995
Thalassodendron_ciliatum	781	261	261	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	0.1696680188179016	0.23079584775086506	0.7387644202930826	0.48240224551051036	0.001
2	0.15703289210796356	0.2342560553633218	0.7582120763265662	0.5678522534916328	0.001
3	0.15008607506752014	0.26089965397923875	0.7749956978144897	0.6070786785177279	0.001
4	0.1459328830242157	0.26332179930795846	0.7809972180612026	0.618893775222464	0.001
5	0.14656734466552734	0.257439446366782	0.7809636417381022	0.6145141477329354	0.001
6	0.14571721851825714	0.2726643598615917	0.7776683649441282	0.6059053020982528	0.001
7	0.14370936155319214	0.2577854671280277	0.7765906414896865	0.6237167782687788	0.001
8	0.14394836127758026	0.2685121107266436	0.7783921362120415	0.6388647657762934	0.001
9	0.14525611698627472	0.255363321799308	0.7855017808506181	0.6320393416159836	0.001
10	0.1438504010438919	0.27058823529411763	0.7851148597100958	0.6486588102502582	0.001
11	0.14624632894992828	0.2629757785467128	0.7877580433216058	0.6417150090850962	0.001
12	0.14636772871017456	0.25432525951557095	0.7837535496084674	0.639315846277914	0.001
13	0.14202021062374115	0.2754325259515571	0.78815800699599	0.6473402316193251	0.001
14	0.1542871743440628	0.25882352941176473	0.7832956146718476	0.6229110923470623	0.001
15	0.14085648953914642	0.2740484429065744	0.7903608334745046	0.6518642313031997	0.001
16	0.14292311668395996	0.27439446366782005	0.7863712916180852	0.6365564755176403	0.001
17	0.13990001380443573	0.2799307958477509	0.7929570954321973	0.6487058751449585	0.001
18	0.13916312158107758	0.2643598615916955	0.7884885951822639	0.6417719350399551	0.001
19	0.13893043994903564	0.26505190311418686	0.7925466885881707	0.655317550465269	0.001
20	0.14047470688819885	0.2778546712802768	0.7910536610255974	0.6519764438062028	0.001
21	0.13958919048309326	0.28096885813148786	0.7944965603502189	0.6586810047475636	0.001
22	0.1409333050251007	0.27231833910034603	0.788903634573398	0.6490927552938025	0.001
23	0.14421699941158295	0.271280276816609	0.791997969886652	0.6546007121389574	0.001
24	0.13853740692138672	0.28512110726643597	0.7920758550626482	0.6473707882350964	0.001
25	0.1393292397260666	0.28269896193771626	0.7918592325630697	0.6582355138095511	0.001
26	0.1404317170381546	0.2771626297577855	0.7923397169025812	0.6563032405626342	0.001
27	0.13901519775390625	0.27854671280276816	0.7912935429967007	0.6372533033632628	0.001
28	0.13772021234035492	0.2709342560553633	0.7872712947561937	0.6456065158774918	0.001
29	0.13838444650173187	0.2771626297577855	0.7939749377663389	0.6443851050256405	0.001
30	0.13830840587615967	0.2640138408304498	0.7942442400202548	0.6491486219528562	0.001
31	0.14057572185993195	0.27612456747404845	0.7925935235222654	0.6310327603311888	0.001
32	0.1388920247554779	0.2833910034602076	0.7971736261865237	0.6698724179590195	0.001
33	0.1613842248916626	0.27024221453287195	0.7971170574103223	0.6497178610524192	0.001
34	0.14063873887062073	0.284083044982699	0.7956378222537918	0.6584094608801518	0.001
35	0.1311163306236267	0.2903114186851211	0.8038767563907229	0.6783097245999954	0.0001
36	0.13003353774547577	0.2996539792387543	0.8076212858821555	0.6793622509723133	0.0001
37	0.1292405128479004	0.3013840830449827	0.8050973873758684	0.6802670830914382	0.0001
38	0.1283079981803894	0.3017301038062284	0.8097280639653521	0.691530412631533	0.0001
39	0.12849266827106476	0.30207612456747407	0.8103419376466646	0.6858958437183065	0.0001
40	0.127731055021286	0.30103806228373703	0.8110104061765693	0.6878904051166272	0.0001
41	0.12703965604305267	0.3079584775086505	0.8139767888452869	0.6967877061883809	0.0001
42	0.12826678156852722	0.30311418685121105	0.8101190476190476	0.6900880377263761	0.0001
43	0.126459002494812	0.3103806228373702	0.813363476513439	0.698512042551847	0.0001
44	0.12749163806438446	0.3110726643598616	0.8125052384544464	0.6928688967130782	0.0001
45	0.1258065104484558	0.3110726643598616	0.8120326165025983	0.6905501287813828	0.0001
46	0.12585382163524628	0.30934256055363324	0.8141467874657136	0.7029042057423242	0.0001
47	0.12546662986278534	0.3096885813148789	0.8159557335755873	0.7047603619742887	0.0001
48	0.1253565400838852	0.3058823529411765	0.8139360013367867	0.6974661734658276	0.0001
49	0.12485454976558685	0.31141868512110726	0.8143054798540942	0.6943635260846596	0.0001
50	0.12531296908855438	0.3103806228373702	0.8110079294406988	0.6932075391775876	0.0001
51	0.12453079223632812	0.31384083044982697	0.814743856726364	0.6991762751520911	0.0001
52	0.12411220371723175	0.31557093425605537	0.8162597533350113	0.7046749806541895	0.0001
53	0.12500016391277313	0.30899653979238756	0.8145090681676048	0.6997083670176113	0.0001
54	0.12449914962053299	0.31453287197231833	0.8163877092180865	0.704905692622788	0.0001
55	0.1243244931101799	0.3134948096885813	0.8149173588388289	0.7015504397667252	0.0001
56	0.12355821579694748	0.3107266435986159	0.8157839909996251	0.7057673948207344	0.0001
57	0.12367301434278488	0.3134948096885813	0.8154531696653775	0.7048308334487045	0.0001
58	0.12418670952320099	0.31418685121107265	0.8152379352345752	0.702233816965674	0.0001
59	0.12396683543920517	0.3169550173010381	0.818200535309862	0.7033190719353933	0.0001
60	0.12383627891540527	0.31591695501730105	0.8169142667441281	0.707100255674436	0.0001
61	0.12376978993415833	0.3134948096885813	0.8161918288913026	0.7057906494932572	0.0001
62	0.12311123311519623	0.3176470588235294	0.8186829126776908	0.710084215624648	0.0001
63	0.12293447554111481	0.3169550173010381	0.8167094358672332	0.7034728877779122	0.0001
64	0.12318814545869827	0.314878892733564	0.8167272423444876	0.7039496449567197	0.0001
65	0.12292256951332092	0.314878892733564	0.8155872667398464	0.7017054590441031	0.0001
66	0.1233849748969078	0.31626297577854673	0.8159070972053971	0.7041658521763916	0.0001
67	0.12293217331171036	0.31626297577854673	0.8193559069459393	0.7115649077332132	0.0001
68	0.12244237214326859	0.32041522491349483	0.8191163512776403	0.7117842251236416	0.0001
69	0.12294486910104752	0.3169550173010381	0.8166987367188154	0.7100473926398085	0.0001
70	0.1227104440331459	0.3176470588235294	0.8194294269317142	0.7073677647969174	0.0001
71	0.12281835079193115	0.3221453287197232	0.8190611783426155	0.7116912727578291	0.0001
72	0.12274286150932312	0.3190311418685121	0.8180748213535098	0.7076888299444815	0.0001
73	0.12224046140909195	0.3197231833910035	0.8208955223880596	0.7115787794437213	0.0001
74	0.12199588865041733	0.3197231833910035	0.8202228504906037	0.7118683173431358	0.0001
75	0.12203484028577805	0.3152249134948097	0.8189331329827199	0.712534792002447	0.0001
76	0.12196851521730423	0.31626297577854673	0.8175121379541269	0.7098431444159636	0.0001
77	0.12242470681667328	0.31418685121107265	0.8165241925295827	0.7095722503090334	0.0001
78	0.12182802706956863	0.3197231833910035	0.8182161594963274	0.7104246992338475	0.0001
79	0.12361899018287659	0.31868512110726643	0.819422117802234	0.7041915269998454	0.0001
80	0.12182050198316574	0.32110726643598614	0.8200709161375442	0.7215166575729017	0.0001
81	0.1222558245062828	0.3190311418685121	0.8196857594147477	0.7142296348553671	0.0001
82	0.12162397801876068	0.3193771626297578	0.8217637573903337	0.7142487132976763	0.0001
83	0.12116336822509766	0.31591695501730105	0.8199182584035366	0.7157416946786116	0.0001
84	0.12182576209306717	0.31591695501730105	0.8160822656776805	0.7005118683471125	0.0001
85	0.12157247215509415	0.32006920415224915	0.8195705855271419	0.7112117400824071	0.0001
86	0.12132923305034637	0.32076124567474046	0.8199150778453084	0.7130384438706758	0.0001
87	0.1209079846739769	0.3245674740484429	0.8217034910331626	0.7196634291835953	0.0001
88	0.12118621915578842	0.32110726643598614	0.8202653799758746	0.7133004794787455	0.0001
89	0.12148680537939072	0.3197231833910035	0.8190659042530991	0.7130024863292826	0.0001
90	0.12146373093128204	0.32110726643598614	0.8195863233431828	0.7138227751305745	0.0001
91	0.12076817452907562	0.3245674740484429	0.8219622861424468	0.7201706813878868	0.0001
92	0.12100402265787125	0.3190311418685121	0.8212632275132274	0.7167137675234206	0.0001
93	0.12075362354516983	0.3169550173010381	0.8210482355397788	0.7168209309560042	0.0001
94	0.12131747603416443	0.3169550173010381	0.8207492316637593	0.7129564562249288	0.0001
95	0.12093706429004669	0.31626297577854673	0.8198186907298326	0.7107923514896998	0.0001
96	0.12080572545528412	0.31833910034602075	0.8205149675528575	0.7142082592350744	0.0001
97	0.12148387730121613	0.3221453287197232	0.821143001430856	0.7178627573063694	0.0001
98	0.12116367369890213	0.31833910034602075	0.8204551137784446	0.7220163155986198	0.0001
99	0.122224360704422	0.3190311418685121	0.8202876735240007	0.7148002470224453	0.0001
100	0.12044612318277359	0.3221453287197232	0.8234614273331406	0.7237819678241929	1e-05
101	0.12046819180250168	0.3214532871972318	0.8225361531388448	0.7175715394879117	1e-05
102	0.11973976343870163	0.3217993079584775	0.823289155622239	0.725621847387927	1e-05
103	0.12033144384622574	0.3214532871972318	0.8229430576899115	0.7238847645251945	1e-05
104	0.12013406306505203	0.32110726643598614	0.8229179659889403	0.7190340295360067	1e-05
105	0.12038043141365051	0.3190311418685121	0.8231939163498099	0.7188152152796179	1e-05
106	0.12012088298797607	0.3217993079584775	0.8225415657114211	0.7162169221873866	1e-05
107	0.11983397603034973	0.3214532871972318	0.8245079469653298	0.7198857151250498	1e-05
108	0.11995845288038254	0.32041522491349483	0.8219728688661866	0.7150987178757772	1e-05
109	0.11981397867202759	0.3262975778546713	0.8223654125961338	0.719215254545586	1.0000000000000002e-06
110	0.11961892992258072	0.3238754325259516	0.8238556410893139	0.7210699392652612	1.0000000000000002e-06
111	0.11969973891973495	0.3256055363321799	0.8240756093516829	0.7220636143470806	1.0000000000000002e-06
112	0.11957906186580658	0.3235294117647059	0.8237543115987199	0.7203145097521634	1.0000000000000002e-06
113	0.11939564347267151	0.32283737024221454	0.8233045212765958	0.7178685303113534	1.0000000000000002e-06
114	0.12025978416204453	0.32006920415224915	0.8221373411940546	0.7198020940141643	1.0000000000000002e-06
115	0.12016712129116058	0.32283737024221454	0.8237909034780819	0.7216303606521312	1.0000000000000002e-06
116	0.12020432949066162	0.32491349480968856	0.8244099021301093	0.724344791703263	1.0000000000000002e-06
117	0.11986654251813889	0.3231833910034602	0.8214361006872708	0.7166667142633425	1.0000000000000002e-06
118	0.11982504278421402	0.31868512110726643	0.8216139293487359	0.7175073424210379	1.0000000000000002e-06
119	0.11932501941919327	0.3231833910034602	0.8241461080833714	0.724230778846038	1.0000000000000002e-06
120	0.11970102787017822	0.32283737024221454	0.8241621845262722	0.7204544328453095	1.0000000000000002e-06
121	0.11957091093063354	0.3221453287197232	0.8231556948798329	0.7230743210923909	1.0000000000000002e-06
122	0.1201123297214508	0.32041522491349483	0.8219534727471236	0.7196142292960622	1.0000000000000002e-06
123	0.11973411589860916	0.3217993079584775	0.8251494537208823	0.7239831576514587	1.0000000000000002e-06
124	0.11995533108711243	0.3221453287197232	0.8215739769150053	0.7195780687908326	1.0000000000000002e-06
125	0.11960428953170776	0.31868512110726643	0.8227369121377187	0.7221896000631038	1.0000000000000002e-06
126	0.11972622573375702	0.3235294117647059	0.8236807168934618	0.7213769755798427	1.0000000000000002e-07
127	0.11995424330234528	0.3197231833910035	0.8234563813938091	0.7220369847258239	1.0000000000000002e-07
128	0.11965782195329666	0.3245674740484429	0.823190583521162	0.7168496424502305	1.0000000000000002e-07
129	0.11992616206407547	0.3256055363321799	0.8250164690382081	0.7255628468971802	1.0000000000000002e-07
130	0.11971699446439743	0.3217993079584775	0.8222776392352452	0.7171476023644834	1.0000000000000002e-07
131	0.1200244128704071	0.32491349480968856	0.8241812768772743	0.7202861226868075	1.0000000000000002e-07
132	0.11962693929672241	0.3238754325259516	0.8228243226370333	0.7204093851990869	1.0000000000000004e-08

---

CO2 Emissions

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

• Emissions: 0.159156917762164 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