groderg/VitEAU-base-patch16-224-2025_11_06_62095-bs32_freeze

VitEAU is a fine-tuned version of VitEAU-base-patch16-224-2025_11_06_62095-bs32_freeze. It achieves the following results on the test set:

• Loss: 0.1528
• F1 Micro: 0.7688
• F1 Macro: 0.6487
• Accuracy: 0.2470

Class	F1 per class
Acropore_branched	0.8349
Acropore_digitised	0.5292
Acropore_sub_massive	0.2185
Acropore_tabular	0.9141
Algae_assembly	0.7493
Algae_drawn_up	0.4235
Algae_limestone	0.7126
Algae_sodding	0.8284
Atra/Leucospilota	0.7156
Bleached_coral	0.6784
Blurred	0.3837
Dead_coral	0.7263
Fish	0.6443
Homo_sapiens	0.7091
Human_object	0.7234
Living_coral	0.6332
Millepore	0.7222
No_acropore_encrusting	0.5904
No_acropore_foliaceous	0.7593
No_acropore_massive	0.6382
No_acropore_solitary	0.4286
No_acropore_sub_massive	0.6399
Rock	0.8640
Rubble	0.7139
Sand	0.9020
Sea_cucumber	0.7400
Sea_urchins	0.5914
Sponge	0.3936
Syringodium_isoetifolium	0.9479
Thalassodendron_ciliatum	0.9644
Useless	0.9692

---

Model description

VitEAU is a model built on top of VitEAU-base-patch16-224-2025_11_06_62095-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: 53.0
• 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	0.18683579564094543	0.1990	0.7024	0.4995	0.001
2	0.17360050976276398	0.2192	0.7345	0.5567	0.001
3	0.1693931370973587	0.2241	0.7473	0.5935	0.001
4	0.16458247601985931	0.2286	0.7504	0.5903	0.001
5	0.16541151702404022	0.2241	0.7500	0.6139	0.001
6	0.1618417203426361	0.2251	0.7580	0.6095	0.001
7	0.16141444444656372	0.2401	0.7537	0.6068	0.001
8	0.16052865982055664	0.2300	0.7528	0.6108	0.001
9	0.16042891144752502	0.2265	0.7562	0.6178	0.001
10	0.15894213318824768	0.2318	0.7563	0.6113	0.001
11	0.15776531398296356	0.2366	0.7635	0.6315	0.001
12	0.15746773779392242	0.2433	0.7590	0.6083	0.001
13	0.15824778378009796	0.2387	0.7571	0.6113	0.001
14	0.1585422158241272	0.2300	0.7592	0.6085	0.001
15	0.15661202371120453	0.2335	0.7616	0.6234	0.001
16	0.15803998708724976	0.2353	0.7574	0.6194	0.001
17	0.15689316391944885	0.2321	0.7657	0.6339	0.001
18	0.15711474418640137	0.2342	0.7614	0.6361	0.001
19	0.1563279777765274	0.2363	0.7649	0.6359	0.001
20	0.15760594606399536	0.2360	0.7576	0.6212	0.001
21	0.15612910687923431	0.2356	0.7632	0.6245	0.001
22	0.15816429257392883	0.2366	0.7636	0.6285	0.001
23	0.1568743884563446	0.2325	0.7647	0.6355	0.001
24	0.15661121904850006	0.2349	0.7660	0.6260	0.001
25	0.15637530386447906	0.2377	0.7663	0.6406	0.001
26	0.15651093423366547	0.2349	0.7632	0.6326	0.001
27	0.1577683985233307	0.2339	0.7617	0.6330	0.001
28	0.15397362411022186	0.2405	0.7690	0.6386	0.0001
29	0.1538531333208084	0.2436	0.7705	0.6437	0.0001
30	0.1537177711725235	0.2440	0.7691	0.6403	0.0001
31	0.1536877304315567	0.2429	0.7698	0.6347	0.0001
32	0.15373626351356506	0.2429	0.7686	0.6358	0.0001
33	0.15358072519302368	0.2408	0.7687	0.6388	0.0001
34	0.1539021134376526	0.2426	0.7673	0.6414	0.0001
35	0.15349489450454712	0.2401	0.7704	0.6410	0.0001
36	0.15370164811611176	0.2405	0.7687	0.6409	0.0001
37	0.153493732213974	0.2412	0.7689	0.6386	0.0001
38	0.1535225659608841	0.2447	0.7688	0.6294	0.0001
39	0.15350276231765747	0.2405	0.7691	0.6387	0.0001
40	0.15352925658226013	0.2422	0.7692	0.6383	0.0001
41	0.153408482670784	0.2412	0.7696	0.6421	0.0001
42	0.15371404588222504	0.2422	0.7672	0.6392	0.0001
43	0.15338563919067383	0.2429	0.7684	0.6381	0.0001
44	0.1535147726535797	0.2405	0.7665	0.6380	0.0001
45	0.15340228378772736	0.2415	0.7701	0.6390	0.0001
46	0.15359722077846527	0.2422	0.7693	0.6362	0.0001
47	0.1536189466714859	0.2443	0.7690	0.6428	0.0001
48	0.1537674516439438	0.2426	0.7670	0.6338	0.0001
49	0.15370245277881622	0.2405	0.7689	0.6390	0.0001
50	0.15347224473953247	0.2415	0.7684	0.6382	1e-05
51	0.1534317433834076	0.2415	0.7685	0.6397	1e-05
52	0.1534072905778885	0.2422	0.7684	0.6395	1e-05
53	0.15339982509613037	0.2426	0.7687	0.6380	1e-05

---

Framework Versions

• Transformers: 4.56.0.dev0
• Pytorch: 2.6.0+cu124
• Datasets: 3.0.2
• Tokenizers: 0.21.0