DinoVdo is a fine-tuned version of DinoVdo-large-2025_01_27_45863-bs32_freeze. It achieves the following results on the test set:

Loss: 0.1236
F1 Micro: 0.8136
F1 Macro: 0.7060
Accuracy: 0.3071

Class	F1 per class
Acropore_branched	0.9015
Acropore_digitised	0.6437
Acropore_sub_massive	0.3853
Acropore_tabular	0.9293
Algae_assembly	0.7615
Algae_drawn_up	0.4765
Algae_limestone	0.7542
Algae_sodding	0.8600
Atra/Leucospilota	0.8369
Bleached_coral	0.6405
Blurred	0.6294
Dead_coral	0.7340
Fish	0.7477
Homo_sapiens	0.7788
Human_object	0.7629
Living_coral	0.6290
Millepore	0.8251
No_acropore_encrusting	0.6364
No_acropore_foliaceous	0.8077
No_acropore_massive	0.7208
No_acropore_solitary	0.4468
No_acropore_sub_massive	0.6970
Rock	0.8818
Rubble	0.7686
Sand	0.9235
Sea_cucumber	0.8234
Sea_urchins	0.7079
Sponge	0.3861
Syringodium_isoetifolium	0.9720
Thalassodendron_ciliatum	0.9886
Useless	0.9745

Model description

DinoVdo is a model built on top of DinoVdo-large-2025_01_27_45863-bs32_freeze 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
Acropore_branched	1480	469	459	2408
Acropore_digitised	571	156	161	888
Acropore_sub_massive	150	52	41	243
Acropore_tabular	999	292	298	1589
Algae_assembly	2554	842	842	4238
Algae_drawn_up	367	130	123	620
Algae_limestone	1651	562	559	2772
Algae_sodding	3142	994	981	5117
Atra/Leucospilota	1084	349	359	1792
Bleached_coral	219	69	72	360
Blurred	191	68	61	320
Dead_coral	1980	648	636	3264
Fish	2018	661	642	3321
Homo_sapiens	161	63	58	282
Human_object	156	55	59	270
Living_coral	397	151	153	701
Millepore	386	127	124	637
No_acropore_encrusting	442	141	142	725
No_acropore_foliaceous	204	47	35	286
No_acropore_massive	1030	341	334	1705
No_acropore_solitary	202	55	46	303
No_acropore_sub_massive	1402	428	426	2256
Rock	4481	1495	1481	7457
Rubble	3092	1015	1016	5123
Sand	5839	1945	1935	9719
Sea_cucumber	1407	437	450	2294
Sea_urchins	328	110	107	545
Sponge	267	98	105	470
Syringodium_isoetifolium	1213	392	390	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: 91.0
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	0.16775080561637878	0.2293	0.7433	0.5138	0.001
2	0.15366077423095703	0.2436	0.7614	0.5771	0.001
3	0.14831368625164032	0.2415	0.7763	0.6194	0.001
4	0.14640773832798004	0.2555	0.7808	0.6276	0.001
5	0.14515382051467896	0.2520	0.7788	0.6421	0.001
6	0.14404040575027466	0.2579	0.7802	0.6147	0.001
7	0.14518576860427856	0.2503	0.7787	0.6141	0.001
8	0.14463570713996887	0.2534	0.7776	0.6193	0.001
9	0.14786836504936218	0.2454	0.7813	0.6363	0.001
10	0.14251072704792023	0.2607	0.7866	0.6366	0.001
11	0.14535894989967346	0.2618	0.7908	0.6566	0.001
12	0.14042578637599945	0.2600	0.7895	0.6439	0.001
13	0.1413952261209488	0.2531	0.7883	0.6490	0.001
14	0.14056158065795898	0.2649	0.7894	0.6348	0.001
15	0.13873133063316345	0.2632	0.7906	0.6470	0.001
16	0.14008578658103943	0.2604	0.7858	0.6331	0.001
17	0.13811993598937988	0.2520	0.7955	0.6651	0.001
18	0.13870170712471008	0.2702	0.7914	0.6498	0.001
19	0.1373777985572815	0.2632	0.7940	0.6356	0.001
20	0.13864819705486298	0.2551	0.7850	0.6393	0.001
21	0.13566707074642181	0.2646	0.7943	0.6496	0.001
22	0.1371580958366394	0.2723	0.7972	0.6400	0.001
23	0.13614478707313538	0.2604	0.7938	0.6595	0.001
24	0.1362675279378891	0.2649	0.7954	0.6418	0.001
25	0.1358671337366104	0.2733	0.7963	0.6500	0.001
26	0.13484793901443481	0.2691	0.7984	0.6555	0.001
27	0.13669784367084503	0.2688	0.7944	0.6535	0.001
28	0.13569706678390503	0.2677	0.7923	0.6398	0.001
29	0.14052562415599823	0.2639	0.7924	0.6637	0.001
30	0.13725879788398743	0.2723	0.7875	0.6405	0.001
31	0.13544805347919464	0.2719	0.7986	0.6562	0.001
32	0.13693773746490479	0.2649	0.7930	0.6463	0.001
33	0.13195939362049103	0.2747	0.8015	0.6660	0.0001
34	0.1299898624420166	0.2834	0.8043	0.6756	0.0001
35	0.12946264445781708	0.2810	0.8066	0.6772	0.0001
36	0.13086125254631042	0.2817	0.8046	0.6795	0.0001
37	0.1278763711452484	0.2824	0.8054	0.6792	0.0001
38	0.12904110550880432	0.2848	0.8078	0.6814	0.0001
39	0.12716704607009888	0.2939	0.8116	0.6833	0.0001
40	0.1293308585882187	0.2908	0.8116	0.6879	0.0001
41	0.12695887684822083	0.2918	0.8089	0.6864	0.0001
42	0.12624548375606537	0.2914	0.8109	0.6837	0.0001
43	0.1261172592639923	0.2949	0.8123	0.6984	0.0001
44	0.12830273807048798	0.2935	0.8106	0.6834	0.0001
45	0.12624593079090118	0.2932	0.8113	0.7010	0.0001
46	0.12462077289819717	0.2960	0.8147	0.6964	0.0001
47	0.12529432773590088	0.2988	0.8126	0.6923	0.0001
48	0.12631145119667053	0.2977	0.8133	0.6954	0.0001
49	0.1252526491880417	0.3037	0.8158	0.6952	0.0001
50	0.12632089853286743	0.3005	0.8136	0.7008	0.0001
51	0.1246422603726387	0.3009	0.8158	0.7019	0.0001
52	0.12534211575984955	0.2911	0.8092	0.6949	0.0001
53	0.12436465919017792	0.3023	0.8154	0.7019	1e-05
54	0.12488020956516266	0.3009	0.8154	0.7040	1e-05
55	0.12366042286157608	0.3005	0.8144	0.6998	1e-05
56	0.12352865934371948	0.3033	0.8168	0.7004	1e-05
57	0.1239086389541626	0.3030	0.8157	0.7002	1e-05
58	0.12343526631593704	0.3026	0.8157	0.6995	1e-05
59	0.12345146387815475	0.3047	0.8150	0.7012	1e-05
60	0.1239377036690712	0.2981	0.8128	0.6932	1e-05
61	0.12398885935544968	0.3009	0.8174	0.7076	1e-05
62	0.12334412336349487	0.3019	0.8152	0.7032	1e-05
63	0.12325507402420044	0.3023	0.8158	0.7023	1e-05
64	0.12346883863210678	0.3047	0.8152	0.6999	1e-05
65	0.12324482202529907	0.2977	0.8145	0.7001	1e-05
66	0.12292143702507019	0.3012	0.8145	0.7004	1e-05
67	0.12375594675540924	0.3016	0.8159	0.6993	1e-05
68	0.1228519007563591	0.2998	0.8176	0.7039	1e-05
69	0.12302352488040924	0.3058	0.8157	0.7006	1e-05
70	0.12284138053655624	0.3040	0.8170	0.7009	1e-05
71	0.12295401096343994	0.3019	0.8158	0.7043	1e-05
72	0.123215451836586	0.3016	0.8171	0.7025	1e-05
73	0.12291014939546585	0.3054	0.8174	0.7049	1e-05
74	0.12304174154996872	0.3009	0.8141	0.6942	1e-05
75	0.1232200339436531	0.3033	0.8161	0.7001	1.0000000000000002e-06
76	0.12267689406871796	0.3058	0.8171	0.7020	1.0000000000000002e-06
77	0.12284990400075912	0.3079	0.8191	0.7060	1.0000000000000002e-06
78	0.123690165579319	0.3019	0.8166	0.7072	1.0000000000000002e-06
79	0.12329532951116562	0.3047	0.8156	0.6992	1.0000000000000002e-06
80	0.12325812131166458	0.3026	0.8172	0.6994	1.0000000000000002e-06
81	0.12240613251924515	0.3054	0.8176	0.7037	1.0000000000000002e-06
82	0.12270382046699524	0.3002	0.8151	0.6972	1.0000000000000002e-06
83	0.12315402179956436	0.2995	0.8146	0.6939	1.0000000000000002e-06
84	0.1225922629237175	0.3026	0.8177	0.7017	1.0000000000000002e-06
85	0.1230376735329628	0.3072	0.8181	0.7059	1.0000000000000002e-06
86	0.12335028499364853	0.3033	0.8168	0.7059	1.0000000000000002e-06
87	0.12355341017246246	0.3005	0.8143	0.6944	1.0000000000000002e-06
88	0.12264065444469452	0.3079	0.8186	0.7029	1.0000000000000002e-07
89	0.1229795441031456	0.3075	0.8179	0.7084	1.0000000000000002e-07
90	0.12317965924739838	0.3037	0.8195	0.7105	1.0000000000000002e-07
91	0.12267619371414185	0.3019	0.8155	0.6972	1.0000000000000002e-07

Framework Versions

Transformers: 4.48.0
Pytorch: 2.5.1+cu124
Datasets: 3.0.2
Tokenizers: 0.21.0

groderg
/

DinoVdo-large-2025_01_27_45863-bs32_freeze