lombardata/DinoVdeau_Aina-large-2024_06_24-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: 0.1203
• F1 Micro: 0.8227
• F1 Macro: 0.7039
• Roc Auc: 0.8807
• Accuracy: 0.3133

---

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	1476	470	462	2408
Acropore_digitised	564	168	156	888
Acropore_sub_massive	150	49	44	243
Acropore_tabular	1000	293	296	1589
Algae_assembly	2543	850	845	4238
Algae_drawn_up	366	127	127	620
Algae_limestone	1653	563	556	2772
Algae_sodding	3145	986	986	5117
Atra/Leucospilota	1083	346	363	1792
Bleached_coral	220	70	70	360
Blurred	191	64	65	320
Dead_coral	1979	640	645	3264
Fish	2021	639	661	3321
Homo_sapiens	161	54	67	282
Human_object	156	55	59	270
Living_coral	149	48	53	250
Millepore	388	126	123	637
No_acropore_encrusting	442	133	150	725
No_acropore_foliaceous	202	43	41	286
No_acropore_massive	1030	339	336	1705
No_acropore_solitary	203	52	48	303
No_acropore_sub_massive	1404	426	426	2256
Rock	4480	1485	1492	7457
Rubble	3090	1016	1017	5123
Sand	5838	1950	1931	9719
Sea_cucumber	1410	437	447	2294
Sea_urchins	328	110	107	545
Sponge	266	98	106	470
Syringodium_isoetifolium	1213	391	391	1995
Thalassodendron_ciliatum	782	261	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	0.16537430882453918	0.23620689655172414	0.75122077697593	0.5257215325423705	0.001
2	0.15053147077560425	0.2537931034482759	0.7619681441057722	0.5537959703923824	0.001
3	0.14628355205059052	0.25482758620689655	0.77513838643595	0.6070050775650959	0.001
4	0.14713425934314728	0.2579310344827586	0.7772792612275123	0.5964805335409908	0.001
5	0.14333128929138184	0.2582758620689655	0.7805424020662936	0.6117223172545019	0.001
6	0.14215321838855743	0.24724137931034482	0.7797190818773552	0.6205823700099015	0.001
7	0.14141972362995148	0.2613793103448276	0.7793963712491276	0.605852706514734	0.001
8	0.1407594084739685	0.25724137931034485	0.7801250929449329	0.6010143467083487	0.001
9	0.143410786986351	0.26448275862068965	0.7832633810739476	0.6180438041585689	0.001
10	0.1399490237236023	0.2655172413793103	0.7821545157780196	0.6118735982260125	0.001
11	0.13849495351314545	0.2775862068965517	0.7860705147664687	0.6206653699553065	0.001
12	0.14007824659347534	0.26275862068965516	0.7848255459845124	0.6298613428358807	0.001
13	0.14162009954452515	0.2596551724137931	0.78046223076587	0.5926571402537333	0.001
14	0.1416475772857666	0.2603448275862069	0.7873954651263039	0.6320114068117629	0.001
15	0.14067792892456055	0.26862068965517244	0.786346178742502	0.6411366319108716	0.001
16	0.1390170454978943	0.26448275862068965	0.7855582680200437	0.633472653157331	0.001
17	0.20845118165016174	0.2520689655172414	0.7772796517954298	0.6345389150334263	0.001
18	0.13100862503051758	0.2872413793103448	0.8010254219183934	0.649970483872118	0.0001
19	0.12990793585777283	0.2824137931034483	0.8003255096796299	0.6536286738545771	0.0001
20	0.13029856979846954	0.28	0.7984633314628566	0.6483260923446458	0.0001
21	0.13044901192188263	0.28379310344827585	0.8020095323972691	0.654053611101578	0.0001
22	0.12825541198253632	0.2910344827586207	0.8026632302405498	0.6578875874990527	0.0001
23	0.13168147206306458	0.28448275862068967	0.8057304652833539	0.6706395135194185	0.0001
24	0.12813876569271088	0.2872413793103448	0.8050271739130435	0.6666793407123319	0.0001
25	0.12696321308612823	0.29517241379310344	0.806903899424675	0.6606483279349962	0.0001
26	0.12607041001319885	0.2944827586206897	0.8066947744985308	0.6716030595890601	0.0001
27	0.12575165927410126	0.2982758620689655	0.8081158929176551	0.6792095054699875	0.0001
28	0.12551021575927734	0.29724137931034483	0.8081711898028954	0.6761647334345041	0.0001
29	0.1253511607646942	0.3003448275862069	0.8078763027507261	0.6746902853962256	0.0001
30	0.1245574951171875	0.29758620689655174	0.8097573349116428	0.6824275599031195	0.0001
31	0.12443281710147858	0.29551724137931035	0.8097552728348619	0.6748384401571104	0.0001
32	0.12430143356323242	0.2989655172413793	0.8074972057432723	0.6788810820138772	0.0001
33	0.12436163425445557	0.2989655172413793	0.8079634297432392	0.684019462951756	0.0001
34	0.12389868497848511	0.3041379310344828	0.8125210650488709	0.6818655593141587	0.0001
35	0.12360044568777084	0.3003448275862069	0.8092445966188744	0.6851516259648143	0.0001
36	0.12417938560247421	0.30517241379310345	0.8093191196698761	0.6809632915243122	0.0001
37	0.12359953671693802	0.3086206896551724	0.8127611090011393	0.6894163666277628	0.0001
38	0.12378468364477158	0.3024137931034483	0.8112101910828026	0.678883648067591	0.0001
39	0.12328560650348663	0.296551724137931	0.8094666438003773	0.6886606577035111	0.0001
40	0.12297580391168594	0.30517241379310345	0.8129929607327622	0.6831509268054475	0.0001
41	0.12383999675512314	0.30689655172413793	0.8124894798855412	0.6929323528561243	0.0001
42	0.12897151708602905	0.306551724137931	0.8110408765276022	0.6896999318804494	0.0001
43	0.12281204015016556	0.30689655172413793	0.8120122158126908	0.6810413750188539	0.0001
44	0.12265903502702713	0.30379310344827587	0.8126684636118598	0.6878141971518951	0.0001
45	0.12264284491539001	0.30310344827586205	0.8148428944486564	0.6921179562112092	0.0001
46	0.12202188372612	0.30551724137931036	0.8161854583772392	0.6940662840133098	0.0001
47	0.12206536531448364	0.30689655172413793	0.8159023979806478	0.6903272160897365	0.0001
48	0.12254682928323746	0.30689655172413793	0.8158389373877448	0.6941956802093657	0.0001
49	0.122016541659832	0.3103448275862069	0.8163869276624529	0.6878662126822556	0.0001
50	0.12227314710617065	0.30448275862068963	0.8150930654623728	0.6933030178910714	0.0001
51	0.12288431823253632	0.30344827586206896	0.8134800900939186	0.6865806997602928	0.0001
52	0.12275662273168564	0.30689655172413793	0.8149179428764292	0.6986468586488102	0.0001
53	0.12184727936983109	0.30896551724137933	0.8141080252172431	0.6957275594183393	1e-05
54	0.12211962789297104	0.31275862068965515	0.8174140697576219	0.7024180868786355	1e-05
55	0.12110267579555511	0.31551724137931036	0.8178746409866532	0.6996137662260781	1e-05
56	0.12095417082309723	0.31275862068965515	0.8165024105123939	0.6980108427132969	1e-05
57	0.12222925573587418	0.3103448275862069	0.8192206270143568	0.7037125040359777	1e-05
58	0.12082309275865555	0.31275862068965515	0.8180280737358363	0.6995037037112045	1e-05
59	0.12129196524620056	0.3175862068965517	0.8200861312037463	0.7076329420605563	1e-05
60	0.12102526426315308	0.3217241379310345	0.821524064171123	0.7008604234504021	1e-05
61	0.12114070355892181	0.3137931034482759	0.817277154351048	0.6978339500045	1e-05
62	0.12093810737133026	0.3137931034482759	0.8185533133925939	0.6968007071200569	1e-05
63	0.12108375132083893	0.3210344827586207	0.8223678752177881	0.7054676006786412	1e-05
64	0.12043146789073944	0.31344827586206897	0.819466644233196	0.7009547135020986	1e-05
65	0.12090969830751419	0.3137931034482759	0.818305546104117	0.6968840725133606	1e-05
66	0.12051945179700851	0.31896551724137934	0.8194250391633855	0.6975519000015916	1e-05
67	0.12076902389526367	0.31413793103448273	0.8214049039098741	0.708879022632045	1e-05
68	0.12054721266031265	0.31689655172413794	0.8205407671347726	0.7070342583250885	1e-05
69	0.12040510773658752	0.31551724137931036	0.8180851516311514	0.7010420184147608	1e-05
70	0.120860755443573	0.31517241379310346	0.8179352432615697	0.6954780619203056	1e-05
71	0.12059099972248077	0.31655172413793103	0.8197053484739752	0.6994719002989627	1e-05
72	0.12049694359302521	0.31896551724137934	0.8194965636463296	0.7006194038535121	1e-05
73	0.12056200951337814	0.3162068965517241	0.8208179684880993	0.7079775830960061	1e-05
74	0.12031462788581848	0.3196551724137931	0.8205321600670438	0.7068983657659407	1e-05
75	0.12019394338130951	0.31413793103448273	0.817958492164337	0.6999797093939304	1e-05
76	0.1204519048333168	0.32	0.8205149710643294	0.706559892743112	1e-05
77	0.12027745693922043	0.3220689655172414	0.8210323606982878	0.701432094679393	1e-05
78	0.12055061757564545	0.3193103448275862	0.8230384663058299	0.7135698196385628	1e-05
79	0.12022813409566879	0.3162068965517241	0.8186892177589853	0.7012593226901207	1e-05
80	0.12015263736248016	0.3179310344827586	0.8199269122526988	0.7055748052257645	1e-05
81	0.12015749514102936	0.31551724137931036	0.8193390659248019	0.7014748515924606	1e-05
82	0.1198667511343956	0.3179310344827586	0.8195974173946069	0.7027863018420247	1e-05
83	0.12045711278915405	0.31655172413793103	0.8191844496091273	0.7007992964828046	1e-05
84	0.1201021820306778	0.3162068965517241	0.8215823128683077	0.7062590552459047	1e-05
85	0.12025153636932373	0.31896551724137934	0.820927903871829	0.706433206208696	1e-05
86	0.12055826932191849	0.31689655172413794	0.8230132450331127	0.7086180348508059	1e-05
87	0.12015349417924881	0.3158620689655172	0.8196555217831814	0.7024093727709093	1e-05
88	0.11993622779846191	0.31724137931034485	0.8209702528716287	0.7080089839961642	1e-05
89	0.12008728086948395	0.3162068965517241	0.8190049077678118	0.7041860231221739	1.0000000000000002e-06
90	0.12053098529577255	0.3162068965517241	0.8202280281038328	0.7018847669788639	1.0000000000000002e-06
91	0.12054412811994553	0.31344827586206897	0.8224360827722936	0.7079497429954984	1.0000000000000002e-06
92	0.12047005444765091	0.31862068965517243	0.8184675501189262	0.6987852059619706	1.0000000000000002e-06

---

CO2 Emissions

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

• Emissions: 1.4006159569205738 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