lombardata/DinoVdeau_Aina-large-2024_06_12-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: 0.1378
• F1 Micro: 0.8118
• F1 Macro: 0.5888
• Roc Auc: 0.8738
• Accuracy: 0.5906

---

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
Acr	509	170	170	849
Ech	149	55	49	253
Gal	149	49	52	250
Mtp	278	93	92	463
Poc	166	54	60	280
Por	265	88	88	441
ALGAE	1221	407	407	2035
RDC	185	65	69	319
SG	1388	463	462	2313
P	198	66	66	330
R	1106	368	369	1843
S	2178	726	726	3630
UNK	132	44	44	220

---

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	0.25564512610435486	0.5342987804878049	0.7703523693803159	0.42604577042963465	0.001
2	0.1855568140745163	0.5449695121951219	0.7607875994972768	0.40573815087342935	0.001
3	0.15800759196281433	0.5335365853658537	0.7772908366533864	0.47696929852236075	0.001
4	0.15843382477760315	0.5274390243902439	0.7700247729149464	0.39442465091480916	0.001
5	0.15615205466747284	0.5442073170731707	0.761375774407178	0.3747774830573479	0.001
6	0.1484147012233734	0.5548780487804879	0.7822349570200573	0.4447304719957427	0.001
7	0.1444726288318634	0.5647865853658537	0.7904456041750301	0.5284590192832462	0.001
8	0.14579781889915466	0.5487804878048781	0.7778469197261979	0.48048557941722936	0.001
9	0.1419043242931366	0.5655487804878049	0.7989700930877401	0.4775190765596756	0.001
10	0.14263293147087097	0.5663109756097561	0.7916750858759345	0.5507171237721333	0.001
11	0.1459268480539322	0.555640243902439	0.7765845441145505	0.43142510371678144	0.001
12	0.14237171411514282	0.5625	0.7899022801302932	0.47044459599035204	0.001
13	0.14195148646831512	0.5586890243902439	0.7957159857199523	0.5304167061533165	0.001
14	0.14145776629447937	0.5685975609756098	0.795869737887212	0.5302959882181708	0.001
15	0.14466659724712372	0.5746951219512195	0.7903159622486664	0.5020556893881151	0.001
16	0.15050330758094788	0.5548780487804879	0.7779618889809444	0.44075758480309546	0.001
17	0.15037894248962402	0.5625	0.7849117174959872	0.5071297581342885	0.001
18	0.15835434198379517	0.5632621951219512	0.7868521879411171	0.4937774793717716	0.001
19	0.13946771621704102	0.5678353658536586	0.795441147573197	0.52511737039556	0.001
20	0.1404852569103241	0.5678353658536586	0.8031007751937984	0.5902607630639873	0.001
21	0.14341644942760468	0.5640243902439024	0.7965933848286789	0.4816903721736541	0.001
22	0.14662735164165497	0.5510670731707317	0.7923046721633294	0.5288404087153705	0.001
23	0.14562036097049713	0.5746951219512195	0.7918968692449356	0.4974177137762122	0.001
24	0.13980671763420105	0.5586890243902439	0.7888934258881176	0.5008214078450646	0.001
25	0.13920389115810394	0.5807926829268293	0.8018232263178755	0.5881062115996991	0.001
26	0.14581459760665894	0.5846036585365854	0.8024120603015076	0.5378532463138629	0.001
27	0.1388114094734192	0.5716463414634146	0.7989535117729925	0.530975959272302	0.001
28	0.14750176668167114	0.5647865853658537	0.7952771662997797	0.4926175534546719	0.001
29	0.14280082285404205	0.5754573170731707	0.7916238965304866	0.4730643714031951	0.001
30	0.14457112550735474	0.5647865853658537	0.7960474308300396	0.5269721557360777	0.001
31	0.2518298327922821	0.555640243902439	0.7859719438877755	0.5163333467519122	0.001
32	0.13625293970108032	0.573170731707317	0.7984836392657622	0.5223572225800281	0.001
33	0.14134813845157623	0.583079268292683	0.797995991983968	0.5094906036573911	0.001
34	0.13918223977088928	0.5617378048780488	0.7939271255060729	0.539855430008089	0.001
35	0.14003774523735046	0.5876524390243902	0.8020833333333333	0.48732799827507844	0.001
36	0.15099692344665527	0.5853658536585366	0.8016096579476861	0.514951340134722	0.001
37	0.14428909122943878	0.5571646341463414	0.7900541407659917	0.48501044595638915	0.001
38	0.14405055344104767	0.5708841463414634	0.7945869521308826	0.5062759605333714	0.001
39	0.13543353974819183	0.5891768292682927	0.8024193548387097	0.535367946456459	0.0001
40	0.1358059197664261	0.59375	0.8035498184751916	0.5341458086147945	0.0001
41	0.13504844903945923	0.5945121951219512	0.8055001992825829	0.5376190537551451	0.0001
42	0.13691695034503937	0.5891768292682927	0.8035073734555599	0.5479834256786992	0.0001
43	0.13574542105197906	0.586890243902439	0.8031840796019901	0.5561858306297028	0.0001
44	0.13488240540027618	0.5891768292682927	0.8039920159680638	0.5433240630212841	0.0001
45	0.1361350268125534	0.5815548780487805	0.8024593415311385	0.5564064426334538	0.0001
46	0.1349516659975052	0.59375	0.8072669826224328	0.5738450359977211	0.0001
47	0.13875022530555725	0.586890243902439	0.8035892323030908	0.5398873319319681	0.0001
48	0.1370573788881302	0.586890243902439	0.8029458598726115	0.5475297928828294	0.0001
49	0.13690504431724548	0.5769817073170732	0.800396432111001	0.5558247567089973	0.0001
50	0.13561294972896576	0.5853658536585366	0.8055390702274975	0.5774903390899776	0.0001
51	0.13570135831832886	0.5899390243902439	0.8056984566679858	0.5639229458996943	1e-05
52	0.13534915447235107	0.5876524390243902	0.803262383131092	0.5523965140127979	1e-05
53	0.13460643589496613	0.5891768292682927	0.8047619047619048	0.5553165219596603	1e-05
54	0.13505160808563232	0.586890243902439	0.80398406374502	0.5513512052592927	1e-05
55	0.13664333522319794	0.5899390243902439	0.8044295036582955	0.5586243767362515	1e-05
56	0.13582760095596313	0.5876524390243902	0.8056215360253365	0.559616045636613	1e-05
57	0.1356770098209381	0.5891768292682927	0.8046205935072694	0.5592413950447728	1e-05
58	0.13557715713977814	0.586890243902439	0.804201347602061	0.5652648109787017	1e-05
59	0.13481777906417847	0.5884146341463414	0.806522171405846	0.564174198827338	1e-05
60	0.13674791157245636	0.5899390243902439	0.8047030689517736	0.5543815900756193	1.0000000000000002e-06
61	0.13405902683734894	0.5876524390243902	0.8045563549160671	0.5561036196553861	1.0000000000000002e-06
62	0.134480819106102	0.5861280487804879	0.8052360174533915	0.5646122700508593	1.0000000000000002e-06
63	0.13709656894207	0.5899390243902439	0.805958291956306	0.5660740924208384	1.0000000000000002e-06
64	0.13475064933300018	0.5929878048780488	0.8056551174830745	0.5521602551489271	1.0000000000000002e-06
65	0.13584908843040466	0.5899390243902439	0.8048345551812959	0.5594424812390786	1.0000000000000002e-06
66	0.13522611558437347	0.5853658536585366	0.8040500297796307	0.5680592850157375	1.0000000000000002e-06
67	0.1348220705986023	0.5876524390243902	0.8059288537549407	0.5689511664868826	1.0000000000000002e-06
68	0.13475003838539124	0.5853658536585366	0.8046843985708614	0.5680973012829017	1.0000000000000002e-07
69	0.13508079946041107	0.5884146341463414	0.8036036036036036	0.5523522321966495	1.0000000000000002e-07
70	0.13654960691928864	0.5853658536585366	0.8053744319304486	0.567977027563922	1.0000000000000002e-07
71	0.13474920392036438	0.5907012195121951	0.8072885719944544	0.5700988754371263	1.0000000000000002e-07

---

CO2 Emissions

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

• Emissions: 0.2101236289384968 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