DinoVdeau is a fine-tuned version of Kamoulox-large-2024_10_31-batch-size64_freeze_monolabel. It achieves the following results on the test set:

Loss: 0.0494
F1 Micro: 0.7640
F1 Macro: 0.3461
Accuracy: 0.7130

Class	F1 per class
ALGAE	0.7961
Acr	0.7462
Acr_Br	0.3797
Anem	0.6767
CCA	0.2710
Ech	0.3610
Fts	0.3889
Gal	0.4667
Gon	0.2222
Mtp	0.5521
P	0.3615
Poc	0.4367
Por	0.5018
R	0.7153
RDC	0.1781
S	0.8252
SG	0.8504
Sarg	0.6303
Ser	0.3252
Slt	0.4188
Sp	0.4198
Turf	0.6045
UNK	0.3763

Model description

DinoVdeau is a model built on top of Kamoulox-large-2024_10_31-batch-size64_freeze_monolabel 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	test	val	Total
ALGAE	36874	12292	12292	61458
Acr	5358	1787	1786	8931
Acr_Br	123	42	42	207
Anem	235	79	79	393
CCA	918	306	306	1530
Ech	618	206	206	1030
Fts	168	57	57	282
Gal	465	155	155	775
Gon	158	53	53	264
Mtp	2370	791	790	3951
P	2658	887	886	4431
Poc	549	184	183	916
Por	1059	354	353	1766
R	31437	10480	10479	52396
RDC	930	310	310	1550
S	57624	19209	19209	96042
SG	25539	8513	8513	42565
Sarg	285	96	96	477
Ser	261	87	87	435
Slt	2730	911	911	4552
Sp	132	44	44	220
Turf	1395	466	466	2327
UNK	292	98	98	488

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

Number of Epochs: 150.0
Learning Rate: 0.001
Train Batch Size: 64
Eval Batch Size: 64
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
0	N/A	N/A	N/A	N/A	0.001
1	0.8028618097305298	0.7326527412414418	0.7326527412414418	0.2723318326456157	0.001
2	0.8038854002952576	0.7288723192975732	0.7288723192975732	0.290650504066453	0.001
3	0.7705450654029846	0.7408581732025574	0.7408581732025574	0.326985383349658	0.001
4	0.7623223066329956	0.7417118168673019	0.7417118168673019	0.310238611675961	0.001
5	0.7626621127128601	0.7383495061061653	0.7383495061061653	0.3107947240093857	0.001
6	0.7451828122138977	0.7447257016428285	0.7447257016428285	0.34331624846537584	0.001
7	0.7458378672599792	0.7467291510600861	0.7467291510600861	0.3283260141196405	0.001
8	0.7398682832717896	0.7458929286946221	0.7458929286946221	0.3352756381207012	0.001
9	0.7424591779708862	0.7455270814097316	0.7455270814097316	0.329402464136349	0.001
10	0.7364382147789001	0.7474782669291475	0.7474782669291475	0.31573051576045374	0.001
11	0.7368418574333191	0.7465375167680005	0.7465375167680005	0.34419509138343996	0.001
12	0.7442134022712708	0.7428093587219735	0.7428093587219735	0.3321199292959056	0.001
13	0.7384127378463745	0.7479312207104406	0.7479312207104406	0.35283143267744377	0.001
14	0.7464041113853455	0.7463633037751956	0.7463633037751956	0.33455884660313173	0.001
15	0.7394037842750549	0.7446734377449871	0.7446734377449871	0.34277495445998946	0.001
16	0.7397111058235168	0.7478789568125991	0.7478789568125991	0.3506456767847629	0.001
17	0.7110718488693237	0.7554398007003362	0.7554398007003362	0.3747287292827094	0.0001
18	0.7041681408882141	0.7567463981463738	0.7567463981463738	0.3792648315920964	0.0001
19	0.7004917860031128	0.7582446298844968	0.7582446298844968	0.38576725361345504	0.0001
20	0.6942671537399292	0.7602829219003153	0.7602829219003153	0.39339544323315423	0.0001
21	0.6919424533843994	0.7592899078413268	0.7592899078413268	0.3942710537489151	0.0001
22	0.6903713941574097	0.7606313478859253	0.7606313478859253	0.3925331634038099	0.0001
23	0.6874070167541504	0.7607010330830474	0.7607010330830474	0.39534246826429575	0.0001
24	0.6864963173866272	0.7612236720614624	0.7612236720614624	0.3933203620961568	0.0001
25	0.684335470199585	0.7614153063535478	0.7614153063535478	0.402343965759826	0.0001
26	0.6830293536186218	0.7629309593909513	0.7629309593909513	0.4055175650763901	0.0001
27	0.6827249526977539	0.7630703297851954	0.7630703297851954	0.40742696523740624	0.0001
28	0.6805527210235596	0.7629309593909513	0.7629309593909513	0.413578373717749	0.0001
29	0.6796479225158691	0.7625825334053413	0.7625825334053413	0.4138447052049864	0.0001
30	0.6774595379829407	0.7635929687636104	0.7635929687636104	0.41283784117218203	0.0001
31	0.677918553352356	0.7643769272312328	0.7643769272312328	0.4100283180344899	0.0001
32	0.6754601001739502	0.7641504503405864	0.7641504503405864	0.41093585032897456	0.0001
33	0.6749601364135742	0.7645162976254769	0.7645162976254769	0.4185852176848548	0.0001
34	0.6746455430984497	0.7650040940053309	0.7650040940053309	0.41458520697205553	0.0001
35	0.6740487813949585	0.7641852929391474	0.7641852929391474	0.42549626405712787	0.0001
36	0.6740365624427795	0.7646905106182819	0.7646905106182819	0.42042185334833837	0.0001
37	0.6731483936309814	0.7641678716398669	0.7641678716398669	0.4194338951085209	0.0001
38	0.671998918056488	0.7648821449103674	0.7648821449103674	0.42433192329853336	0.0001
39	0.6694707870483398	0.7659274228671974	0.7659274228671974	0.42040362773805245	0.0001
40	0.6693674325942993	0.7665023257434539	0.7665023257434539	0.42052839725843943	0.0001
41	0.6682748198509216	0.7658228950715145	0.7658228950715145	0.4176095837463332	0.0001
42	0.6682831645011902	0.7665371683420149	0.7665371683420149	0.4307432846853069	0.0001
43	0.6695142984390259	0.7664152192470515	0.7664152192470515	0.42637174689527435	0.0001
44	0.669391393661499	0.765840316370795	0.765840316370795	0.42609728385101026	0.0001
45	0.6695447564125061	0.7649518301074895	0.7649518301074895	0.4307801257532618	0.0001
46	0.6653340458869934	0.7673908120067595	0.7673908120067595	0.4362286100546047	0.0001
47	0.6659862995147705	0.7674430759046009	0.7674430759046009	0.43279477858934173	0.0001
48	0.665557861328125	0.7668855943276249	0.7668855943276249	0.43077383038275147	0.0001
49	0.6672787666320801	0.7666242748384174	0.7666242748384174	0.4241681801855841	0.0001
50	0.6661437749862671	0.7662584275535269	0.7662584275535269	0.43151276516162357	0.0001
51	0.6638755798339844	0.7667288026341005	0.7667288026341005	0.4307690989303114	0.0001
52	0.6654694676399231	0.7679134509851745	0.7679134509851745	0.4427799679621408	0.0001
53	0.6643231511116028	0.767234020313235	0.767234020313235	0.4341825403324277	0.0001
54	0.667382001876831	0.7663455340499294	0.7663455340499294	0.4459616186399035	0.0001
55	0.6627440452575684	0.7684709325621505	0.7684709325621505	0.4389385481332223	0.0001
56	0.6627209186553955	0.767094649918991	0.767094649918991	0.43857797557707	0.0001
57	0.6640397310256958	0.7669204369261859	0.7669204369261859	0.43847119749006624	0.0001
58	0.6627684235572815	0.7672862842110765	0.7672862842110765	0.43760916892251167	0.0001
59	0.6614954471588135	0.7679134509851745	0.7679134509851745	0.439932052501894	0.0001
60	0.6633245944976807	0.766990122123308	0.766990122123308	0.44188579219640917	0.0001
61	0.6611309051513672	0.7685754603578335	0.7685754603578335	0.4370683373238057	0.0001
62	0.660831093788147	0.7684360899635895	0.7684360899635895	0.45352172583851785	0.0001
63	0.6621896028518677	0.767791501890211	0.767791501890211	0.44611945476580944	0.0001
64	0.6610415577888489	0.767547603700284	0.767547603700284	0.4439334258360575	0.0001
65	0.6589834690093994	0.7680354000801379	0.7680354000801379	0.434573899819693	0.0001
66	0.6599727272987366	0.76845351126287	0.76845351126287	0.4397010901020239	0.0001
67	0.6572328209877014	0.769045835438407	0.769045835438407	0.44838573955584105	0.0001
68	0.658860445022583	0.7686799881535165	0.7686799881535165	0.4442341389313245	0.0001
69	0.659292995929718	0.7688542011463215	0.7688542011463215	0.43926108990057783	0.0001
70	0.658970832824707	0.7679134509851745	0.7679134509851745	0.4357439201261406	0.0001
71	0.6567061543464661	0.768244455671504	0.768244455671504	0.4432082950234043	0.0001
72	0.65887850522995	0.7681225065765405	0.7681225065765405	0.4369170898714745	0.0001
73	0.6611541509628296	0.7675998675981255	0.7675998675981255	0.4436833314710758	0.0001
74	0.6570971012115479	0.768488353861431	0.768488353861431	0.44906341810873124	0.0001
75	0.6557245254516602	0.768174770474382	0.768174770474382	0.44439364748025234	0.0001
76	0.658838152885437	0.7683489834671869	0.7683489834671869	0.4478607285233603	0.0001
77	0.6572225093841553	0.7686277242556749	0.7686277242556749	0.44887316801028765	0.0001
78	0.6562930941581726	0.768767094649919	0.768767094649919	0.4439839848601264	0.0001
79	0.6564787030220032	0.76810508527726	0.76810508527726	0.4379298766193662	0.0001
80	0.661143958568573	0.768383826065748	0.768383826065748	0.4460529321244195	0.0001
81	0.660437285900116	0.768941307642724	0.768941307642724	0.44750591322776384	0.0001
82	0.6531779766082764	0.7705614884758105	0.7705614884758105	0.4526720456188751	1e-05
83	0.6532895565032959	0.77019564119092	0.77019564119092	0.4489367718812771	1e-05
84	0.6505005359649658	0.7705266458772495	0.7705266458772495	0.45139096558153424	1e-05
85	0.6501905918121338	0.7708053866657375	0.7708053866657375	0.4565625671001629	1e-05
86	0.6507149338722229	0.7707182801693351	0.7707182801693351	0.4559124472677571	1e-05
87	0.6484472751617432	0.770927335760701	0.770927335760701	0.45838476700319497	1e-05
88	0.6496042013168335	0.77054406717653	0.77054406717653	0.4569340367642959	1e-05
89	0.6486304402351379	0.7713977108412745	0.7713977108412745	0.45601319436503734	1e-05
90	0.6490767598152161	0.7711886552499085	0.7711886552499085	0.45742417795749246	1e-05
91	0.6481940746307373	0.7704221180815666	0.7704221180815666	0.45182909085509243	1e-05
92	0.6477252244949341	0.7715545025347991	0.7715545025347991	0.45503456574154166	1e-05
93	0.6489835381507874	0.771206076549189	0.771206076549189	0.4518083431267937	1e-05
94	0.6485304832458496	0.770753122767896	0.770753122767896	0.45105746851856937	1e-05
95	0.647895336151123	0.771659030330482	0.771659030330482	0.45671492126995916	1e-05
96	0.6472702622413635	0.7715022386369575	0.7715022386369575	0.4596959338122155	1e-05
97	0.6460831165313721	0.7713977108412745	0.7713977108412745	0.4625401473178366	1e-05
98	0.6463102698326111	0.7722165119074581	0.7722165119074581	0.45892155771298887	1e-05
99	0.646852433681488	0.77089249316214	0.77089249316214	0.4549241891767946	1e-05
100	0.6456441879272461	0.7723384610024215	0.7723384610024215	0.45970146016594643	1e-05
101	0.6469387412071228	0.7717461368268845	0.7717461368268845	0.4573593202819609	1e-05
102	0.646738588809967	0.7717809794254455	0.7717809794254455	0.4593480600391769	1e-05
103	0.6467755436897278	0.7723036184038605	0.7723036184038605	0.4576617106244536	1e-05
104	0.6456966400146484	0.7725475165937876	0.7725475165937876	0.4578893476938316	1e-05
105	0.6455578804016113	0.7719377711189701	0.7719377711189701	0.45556962818046953	1e-05
106	0.6444206237792969	0.7723384610024215	0.7723384610024215	0.4644160307431161	1e-05
107	0.26552170515060425	0.04797825821849794	0.4910938804941607	0.360721302464235	1e-05
108	0.1419014185667038	0.44983536872179924	0.6693680656054029	0.22462065038139475	1e-05
109	0.07755623757839203	0.6714691381683245	0.7449736568518617	0.2136927959803109	1e-05
110	0.05802077427506447	0.6837163115625163	0.7489470111853911	0.26414762709864326	1e-05
111	0.053473543375730515	0.6935245030574381	0.7547299175391458	0.331603552491686	1e-05
112	0.05167479068040848	0.6998135920976987	0.7585280588776449	0.3483537603081725	1e-05
113	0.05106380954384804	0.7042734447135067	0.7611001027447527	0.3378893620669972	1e-05
114	0.05065497010946274	0.7053535652688978	0.7622047244094489	0.35703913789456687	1e-05
115	0.05039990693330765	0.7117820247034023	0.7647240545893983	0.36428903036337407	1e-05
116	0.05015714839100838	0.7101966864688769	0.7647289615591668	0.3622993891059384	1e-05
117	0.050175271928310394	0.712914409156635	0.7657223847509677	0.36544151863175506	1e-05
118	0.050468478351831436	0.7141687427048309	0.7654782537680462	0.3524192831401073	1e-05
119	0.049900032579898834	0.7127053535652689	0.7658673932788375	0.34416444697858145	1e-05
120	0.049903545528650284	0.7130886221494399	0.7657258505633957	0.35077501544817247	1e-05
121	0.04957958310842514	0.7140816362084285	0.7665756914119359	0.3627670615797559	1e-05
122	0.04973344877362251	0.7163812477134545	0.7672263726699065	0.35293584502475456	1e-05
123	0.04949206858873367	0.7153533910559049	0.7661930650098223	0.36741171960145996	1e-05
124	0.049613192677497864	0.7160676643264055	0.7673595994775795	0.36413807211107735	1e-05
125	0.04959910735487938	0.7124440340760614	0.7658070643240676	0.3509397162428964	1e-05
126	0.0494619682431221	0.7152662845595025	0.7660751240774316	0.37424111866414195	1e-05
127	0.049399666488170624	0.7149178585738925	0.766314294299216	0.35735030768195364	1e-05
128	0.04938925430178642	0.714412640894758	0.7664776721721585	0.36010176970077795	1e-05
129	0.049368634819984436	0.717931743349419	0.7674323253122921	0.3641550142658243	1e-05
130	0.049409620463848114	0.717687845159492	0.7667705923765463	0.3602711408206009	1e-05
131	0.04939533770084381	0.718210484137907	0.7665082507046622	0.3664294602272974	1e-05
132	0.04943186417222023	0.717583317363809	0.7664564319910658	0.3651176446191739	1e-05
133	0.049288176000118256	0.714621696486124	0.7658437005098911	0.36115748131858555	1e-05
134	0.04927274212241173	0.7154753401508684	0.7659699195779215	0.3677607284274943	1e-05
135	0.04929700121283531	0.7189944426055295	0.7674080308866179	0.37226237632641596	1e-05
136	0.049187980592250824	0.7151094928659779	0.766711291239524	0.35969130867283144	1e-05
137	0.0491538941860199	0.7157889235379175	0.7664713487937058	0.3631043131303743	1e-05
138	0.04930136725306511	0.7178446368530165	0.7665450277813434	0.3687935119842292	1e-05
139	0.04927237331867218	0.7182279054371875	0.766155421092079	0.35626916649899915	1e-05
140	0.04918988421559334	0.7197958223724326	0.767376184687937	0.3699733355385332	1e-05
141	0.04920462518930435	0.716468354209857	0.7666041104041745	0.35072596287625124	1e-05
142	0.04919710010290146	0.7194473963868225	0.7669340748803981	0.36600085102264546	1e-05
143	0.04930509999394417	0.7168342014947475	0.765517685242224	0.3673237139632794	1e-05
144	0.0490318201482296	0.7171477848817965	0.7667940015206897	0.3554021435508309	1.0000000000000002e-06
145	0.04918621480464935	0.7201616696573231	0.7677822164123848	0.3711029550898432	1.0000000000000002e-06
146	0.04903709515929222	0.717130363582516	0.7665065530257804	0.368326447075977	1.0000000000000002e-06
147	0.049094948917627335	0.720823679029982	0.768544776459646	0.37476196073915324	1.0000000000000002e-06
148	0.04907181113958359	0.7167296736990645	0.7667018106807243	0.3649988602534311	1.0000000000000002e-06
149	0.04904184117913246	0.718210484137907	0.7671139893046166	0.3787860055208151	1.0000000000000002e-06
150	0.04912904277443886	0.7154404975523074	0.7667920374277589	0.3726446424747262	1.0000000000000002e-06

Framework Versions

Transformers: 4.44.2
Pytorch: 2.4.1+cu121
Datasets: 3.0.0
Tokenizers: 0.19.1

groderg
/

Kamoulox-large-2024_10_31-batch-size64_freeze_monolabel