lombardata/dinov2-large-2024_05_27-_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: 0.1247
• F1 Micro: 0.8153
• F1 Macro: 0.7021
• Roc Auc: 0.8747
• Accuracy: 0.3144

---

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	0.1701383888721466	0.232726022688209	0.7380235658381353	0.4712591871520079	0.001
1.8315018315018317	N/A	N/A	N/A	N/A	0.001
2.0	0.15890291333198547	0.24888277758679958	0.7568708574323469	0.5722657636852799	0.001
3.0	0.15134122967720032	0.24785149535922998	0.7723932964583505	0.6104117366005594	0.001
3.663003663003663	N/A	N/A	N/A	N/A	0.001
4.0	0.15164224803447723	0.24853901684427637	0.7608496532472631	0.599745200497783	0.001
5.0	0.15243493020534515	0.24750773461670678	0.7692371752165224	0.5935106518877853	0.001
5.4945054945054945	N/A	N/A	N/A	N/A	0.001
6.0	0.14673969149589539	0.24269508422138192	0.7718080548414739	0.613788061665736	0.001
7.0	0.1506606936454773	0.2509453420419388	0.7732481363152289	0.6309898748773293	0.001
7.326007326007326	N/A	N/A	N/A	N/A	0.001
8.0	0.14430351555347443	0.2609144035751117	0.7828014555188422	0.6403740520777896	0.001
9.0	0.14617429673671722	0.25128910278446204	0.781416038551835	0.6366498775226099	0.001
9.157509157509157	N/A	N/A	N/A	N/A	0.001
10.0	0.14414818584918976	0.2688209006531454	0.7794745970641737	0.6254472467805158	0.001
10.989010989010989	N/A	N/A	N/A	N/A	0.001
11.0	0.14589238166809082	0.2595393606050189	0.7780349253103302	0.6357434835994208	0.001
12.0	0.14458976686000824	0.2554142316947405	0.7823495795575149	0.638389910481932	0.001
12.820512820512821	N/A	N/A	N/A	N/A	0.001
13.0	0.14142437279224396	0.2561017531797869	0.786284091383703	0.6574365741219022	0.001
14.0	0.1581379920244217	0.24682021313166036	0.7766990291262137	0.6245865910731833	0.001
14.652014652014651	N/A	N/A	N/A	N/A	0.001
15.0	0.1447945237159729	0.2598831213475421	0.7859620485615181	0.6552072842486797	0.001
16.0	0.1438169628381729	0.2605706428325885	0.7853051058530511	0.6495757946819554	0.001
16.483516483516482	N/A	N/A	N/A	N/A	0.001
17.0	0.14359386265277863	0.2506015812994156	0.7824457675812967	0.6310969679900952	0.001
18.0	0.1412857472896576	0.2564455139223101	0.7848311343456975	0.6531950395959965	0.001
18.315018315018314	N/A	N/A	N/A	N/A	0.001
19.0	0.14079046249389648	0.26022688209006534	0.7833830386020918	0.6486819478687708	0.001
20.0	0.14640754461288452	0.26400825025782054	0.7775968460747342	0.6262168341318395	0.001
20.146520146520146	N/A	N/A	N/A	N/A	0.001
21.0	0.1412632316350937	0.2653832932279134	0.7890916719110552	0.6582044080070929	0.001
21.978021978021978	N/A	N/A	N/A	N/A	0.001
22.0	0.14168681204319	0.2543829494671708	0.7871090517954659	0.6586947128782558	0.001
23.0	0.1393543779850006	0.269852182880715	0.7863651704353696	0.6427873985434494	0.001
23.80952380952381	N/A	N/A	N/A	N/A	0.001
24.0	0.14052371680736542	0.2588518391199725	0.7857706852844616	0.6618962794412713	0.001
25.0	0.1392364352941513	0.2653832932279134	0.7897693920335429	0.653320279245233	0.001
25.641025641025642	N/A	N/A	N/A	N/A	0.001
26.0	0.14239099621772766	0.27019594362323823	0.7838044308632545	0.6529431984792132	0.001
27.0	0.1386287957429886	0.2671020969405294	0.7974886125815585	0.6810613208979668	0.001
27.47252747252747	N/A	N/A	N/A	N/A	0.001
28.0	0.15519200265407562	0.2650395324853902	0.7791304347826087	0.6474807711800876	0.001
29.0	0.14190098643302917	0.27019594362323823	0.7913651213762871	0.6550381793679035	0.001
29.304029304029303	N/A	N/A	N/A	N/A	0.001
30.0	0.13986903429031372	0.2767273977311791	0.7857173292428311	0.663185953977854	0.001
31.0	0.13765402138233185	0.27260226882090066	0.7881844380403459	0.6554744698272669	0.001
31.135531135531135	N/A	N/A	N/A	N/A	0.001
32.0	0.13866138458251953	0.2677896184255758	0.7914770376499792	0.6596978272887946	0.001
32.967032967032964	N/A	N/A	N/A	N/A	0.001
33.0	0.13930276036262512	0.2605706428325885	0.7887546855476885	0.6583814800932023	0.001
34.0	0.1374826431274414	0.2763836369886559	0.795303262082937	0.6636727922636001	0.001
34.798534798534796	N/A	N/A	N/A	N/A	0.001
35.0	0.14001137018203735	0.25850807837744927	0.7860775988902434	0.6442491093834092	0.001
36.0	0.13899104297161102	0.26916466139566864	0.7890085033301218	0.6541220211466265	0.001
36.63003663003663	N/A	N/A	N/A	N/A	0.001
37.0	0.14101693034172058	0.2667583361980062	0.788356222091162	0.6602790790864311	0.001
38.0	0.13849563896656036	0.2633207287727741	0.7864065343433915	0.6508514926081754	0.001
38.46153846153846	N/A	N/A	N/A	N/A	0.001
39.0	0.14249388873577118	0.26263320728772777	0.7819844457738655	0.6513021077089046	0.001
40.0	0.1512959599494934	0.2633207287727741	0.7819497946916141	0.6421624481517915	0.001
40.29304029304029	N/A	N/A	N/A	N/A	0.0001
41.0	0.1416281908750534	0.27157098659333107	0.795353889863792	0.6708412782877394	0.0001
42.0	0.13480685651302338	0.2811962873839807	0.8014906832298136	0.6820172839356666	0.0001
42.124542124542124	N/A	N/A	N/A	N/A	0.0001
43.0	0.1342025101184845	0.2756961155036095	0.8014919187733112	0.681931169239128	0.0001
43.956043956043956	N/A	N/A	N/A	N/A	0.0001
44.0	0.1327475756406784	0.2811962873839807	0.8019789631231031	0.683693351140427	0.0001
45.0	0.1318245828151703	0.2811962873839807	0.8049446006284108	0.6900135704395078	0.0001
45.78754578754579	N/A	N/A	N/A	N/A	0.0001
46.0	0.13027183711528778	0.28910278446201443	0.8063969585520062	0.6920134474185277	0.0001
47.0	0.12985946238040924	0.284977655551736	0.8065087538619978	0.6938459582689339	0.0001
47.61904761904762	N/A	N/A	N/A	N/A	0.0001
48.0	0.12981055676937103	0.2853214162942592	0.8031727379553465	0.6917397436201066	0.0001
49.0	0.1301460713148117	0.2839463733241664	0.8081048867699644	0.6980761423122126	0.0001
49.45054945054945	N/A	N/A	N/A	N/A	0.0001
50.0	0.1294524371623993	0.2829150910965968	0.8056895691232739	0.6968263757426811	0.0001
51.0	0.12989668548107147	0.2846338948092128	0.8078541374474054	0.6981227572539419	0.0001
51.282051282051285	N/A	N/A	N/A	N/A	0.0001
52.0	0.13097986578941345	0.284977655551736	0.809621541745341	0.7032059573412642	0.0001
53.0	0.12910524010658264	0.288415262976968	0.8082875892525485	0.6952081515364695	0.0001
53.11355311355312	N/A	N/A	N/A	N/A	0.0001
54.0	0.1276824176311493	0.2860089377793056	0.8055729885778838	0.6914506394370794	0.0001
54.94505494505494	N/A	N/A	N/A	N/A	0.0001
55.0	0.12751279771327972	0.28979030594706084	0.8091508143727464	0.7051415507931676	0.0001
56.0	0.12798655033111572	0.2911653489171537	0.8077718065316246	0.6990943862949641	0.0001
56.776556776556774	N/A	N/A	N/A	N/A	0.0001
57.0	0.1279618740081787	0.29150910965967686	0.8107930240210597	0.7001268142729874	0.0001
58.0	0.1280883550643921	0.290134066689584	0.8108946874106743	0.7039327958876614	0.0001
58.608058608058606	N/A	N/A	N/A	N/A	0.0001
59.0	0.1287168562412262	0.2873839807493984	0.8071845383437488	0.699653006099352	0.0001
60.0	0.1270500272512436	0.28875902371949125	0.8103491168421926	0.7042073996338176	0.0001
60.43956043956044	N/A	N/A	N/A	N/A	0.0001
61.0	0.1269637793302536	0.28944654520453766	0.8072888368788399	0.6994480698947442	0.0001
62.0	0.12639474868774414	0.28979030594706084	0.8124407826982492	0.7105518005302388	0.0001
62.27106227106227	N/A	N/A	N/A	N/A	0.0001
63.0	0.12643341720104218	0.2918528704022001	0.8093336660843524	0.7042257858113937	0.0001
64.0	0.12570597231388092	0.2918528704022001	0.8119739624362535	0.7054117610081568	0.0001
64.1025641025641	N/A	N/A	N/A	N/A	0.0001
65.0	0.12599390745162964	0.29322791337229287	0.8103770839396333	0.7040599127700347	0.0001
65.93406593406593	N/A	N/A	N/A	N/A	0.0001
66.0	0.12674611806869507	0.29769680302509455	0.8141795311606633	0.7083351143800681	0.0001
67.0	0.12676431238651276	0.28979030594706084	0.8090950582963362	0.6998024530144022	0.0001
67.76556776556777	N/A	N/A	N/A	N/A	0.0001
68.0	0.12638631463050842	0.2928841526297697	0.8127327032445482	0.7034736625177254	0.0001
69.0	0.12608103454113007	0.2952904778274321	0.8131967584022379	0.7078892431331377	0.0001
69.59706959706959	N/A	N/A	N/A	N/A	0.0001
70.0	0.12582050263881683	0.29150910965967686	0.8136722606120435	0.7081157868651535	0.0001
71.0	0.12533149123191833	0.2918528704022001	0.8123295595405339	0.7044517956080781	1e-05
71.42857142857143	N/A	N/A	N/A	N/A	1e-05
72.0	0.1258901059627533	0.2966655207975249	0.8159506713723581	0.7099295458861072	1e-05
73.0	0.12526649236679077	0.2949467170849089	0.8159496670343587	0.7116557450872655	1e-05
73.26007326007326	N/A	N/A	N/A	N/A	1e-05
74.0	0.12490212172269821	0.29769680302509455	0.8156100747030249	0.7159515864206437	1e-05
75.0	0.12504002451896667	0.2966655207975249	0.8135426082669078	0.7082306828309269	1e-05
75.0915750915751	N/A	N/A	N/A	N/A	1e-05
76.0	0.12634462118148804	0.2966655207975249	0.8099675513769865	0.6998917153140419	1e-05
76.92307692307692	N/A	N/A	N/A	N/A	1e-05
77.0	0.1249643936753273	0.2966655207975249	0.8142915811088296	0.7104044870773909	1e-05
78.0	0.12509745359420776	0.2939154348573393	0.812339968613199	0.7076718539561497	1e-05
78.75457875457876	N/A	N/A	N/A	N/A	1e-05
79.0	0.12465520948171616	0.29838432451014096	0.8147326016360423	0.7097766100728804	1e-05
80.0	0.12526248395442963	0.2990718459951873	0.8166140393490405	0.7133791911991404	1e-05
80.58608058608058	N/A	N/A	N/A	N/A	1e-05
81.0	0.12510864436626434	0.2952904778274321	0.8121923983622152	0.705898272950067	1e-05
82.0	0.12532733380794525	0.29975936748023374	0.8150326797385622	0.7095032932540235	1e-05
82.41758241758242	N/A	N/A	N/A	N/A	1e-05
83.0	0.12474868446588516	0.29597799931247853	0.815855206584497	0.7124383950303705	1e-05
84.0	0.12511858344078064	0.3007906497078034	0.8175330467926365	0.7138847615465347	1e-05
84.24908424908425	N/A	N/A	N/A	N/A	1e-05
85.0	0.12457013875246048	0.2966655207975249	0.8132141082960754	0.7054571621251418	1e-05
86.0	0.1251869946718216	0.2946029563423857	0.8143732269868025	0.7142702846379808	1e-05
86.08058608058609	N/A	N/A	N/A	N/A	1e-05
87.0	0.12492978572845459	0.2935716741148161	0.8135328455150868	0.7081357577756824	1e-05
87.91208791208791	N/A	N/A	N/A	N/A	1e-05
88.0	0.12513719499111176	0.2990718459951873	0.815831263487927	0.7099379006276698	1e-05
89.0	0.12514576315879822	0.2963217600550017	0.8143914473684211	0.7092910188720426	1e-05
89.74358974358974	N/A	N/A	N/A	N/A	1e-05
90.0	0.1244530975818634	0.2942591955998625	0.8134516195584898	0.7121664381657501	1e-05
91.0	0.12501013278961182	0.2990718459951873	0.8153902768123646	0.7106178930468596	1e-05
91.57509157509158	N/A	N/A	N/A	N/A	1e-05
92.0	0.12525025010108948	0.2973530422825713	0.8163049232398094	0.7140173113811211	1e-05
93.0	0.12471849471330643	0.29872808525266414	0.8148661314641998	0.7129019083206937	1e-05
93.4065934065934	N/A	N/A	N/A	N/A	1e-05
94.0	0.12515641748905182	0.2980405637676177	0.8141884924726748	0.7053935701419592	1e-05
95.0	0.12481416761875153	0.30147817119284975	0.8165906870726147	0.7134995447430972	1e-05
95.23809523809524	N/A	N/A	N/A	N/A	1e-05
96.0	0.12492986023426056	0.2980405637676177	0.8160666176830762	0.7110442004495683	1e-05
97.0	0.12459924072027206	0.30147817119284975	0.8168590473093806	0.7158597011246477	1.0000000000000002e-06
97.06959706959707	N/A	N/A	N/A	N/A	1.0000000000000002e-06
98.0	0.12447398155927658	0.29975936748023374	0.8149457415323906	0.707122866121441	1.0000000000000002e-06
98.9010989010989	N/A	N/A	N/A	N/A	1.0000000000000002e-06
99.0	0.12462905794382095	0.30216569267789617	0.8165748111859562	0.7182970295785608	1.0000000000000002e-06
100.0	0.12463195621967316	0.30147817119284975	0.8161644284310514	0.7136275002413193	1.0000000000000002e-06

---

CO2 Emissions

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

• Emissions: 1.562242452449767 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