drone-DinoVdeau-produttoria_binary-binary is a fine-tuned version of drone-DinoVdeau-produttoria_binary-binary-large-2024_11_03-batch-size64_freeze. It achieves the following results on the test set:

Loss: 0.2854
F1 Micro: 0.8468
F1 Macro: 0.6351
Accuracy: 0.2786

Class	F1 per class
Acropore_branched	0.8084
Acropore_digitised	0.5125
Acropore_tabular	0.3951
Algae	0.9562
Dead_coral	0.7470
Fish	0.6639
Millepore	0.3021
No_acropore_encrusting	0.5923
No_acropore_massive	0.7651
No_acropore_sub_massive	0.6345
Rock	0.9536
Rubble	0.9042
Sand	0.9008

Model description

drone-DinoVdeau-produttoria_binary-binary is a model built on top of drone-DinoVdeau-produttoria_binary-binary-large-2024_11_03-batch-size64_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	1483	522	529	2534
Acropore_digitised	1085	371	362	1818
Acropore_tabular	486	176	178	840
Algae	10340	3441	3461	17242
Dead_coral	3710	1252	1267	6229
Fish	1462	517	515	2494
Millepore	746	282	273	1301
No_acropore_encrusting	1993	751	728	3472
No_acropore_massive	4450	1581	1649	7680
No_acropore_sub_massive	3034	1102	1113	5249
Rock	10225	3429	3445	17099
Rubble	9353	3100	3105	15558
Sand	9271	3101	3132	15504

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

Number of Epochs: 88.0
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
1	0.3235681354999542	0.2630072840790843	0.8262109753225342	0.5774239185038708	0.001
2	0.3146470785140991	0.24115504682622269	0.8378565084377776	0.6199165901601139	0.001
3	0.3090434670448303	0.2554630593132154	0.8398465111582348	0.6043570009634397	0.001
4	0.30735355615615845	0.25624349635796045	0.8348980169243037	0.600278483167516	0.001
5	0.30385810136795044	0.2515608740894901	0.8405948994360434	0.6247746971203368	0.001
6	0.3059956729412079	0.2596253902185224	0.841987466427932	0.6225111439021958	0.001
7	0.3013758361339569	0.28199791883454733	0.8387498056289846	0.5954695621655504	0.001
8	0.30131709575653076	0.2702913631633715	0.8390550208451284	0.5974832028652961	0.001
9	0.30098479986190796	0.28407908428720086	0.8406665130922214	0.5974259992816957	0.001
10	0.30072343349456787	0.27107180020811655	0.8376187886791475	0.5937940362628795	0.001
11	0.3035621643066406	0.277315296566077	0.8348592565387339	0.5761905737205768	0.001
12	0.3012838363647461	0.26742976066597296	0.838466245156027	0.6114755503631268	0.001
13	0.29778778553009033	0.2648283038501561	0.8421213122252433	0.6145726431106396	0.001
14	0.29774588346481323	0.27341311134235174	0.8399742101869762	0.605884177295118	0.001
15	0.29809942841529846	0.2666493236212279	0.8433503513117323	0.6074624445346274	0.001
16	0.29744812846183777	0.27471383975026015	0.8394100355835181	0.5932952143692389	0.001
17	0.2983638644218445	0.2663891779396462	0.8437578624264077	0.6146867059353278	0.001
18	0.3023049235343933	0.2762747138397503	0.8356339535005088	0.5803903225868541	0.001
19	0.2984697222709656	0.2739334027055151	0.8423529411764706	0.6158875389283108	0.001
20	0.29680272936820984	0.28069719042663893	0.8411767731317183	0.5984147849283556	0.001
21	0.30051520466804504	0.2702913631633715	0.8418969323285377	0.6060492619397649	0.001
22	0.29818177223205566	0.27471383975026015	0.8374817746302854	0.580353532272699	0.001
23	0.29393449425697327	0.27809573361082207	0.8436262061960386	0.615237110287355	0.001
24	0.2948347330093384	0.27601456815816855	0.8453232862164007	0.6228721497006335	0.001
25	0.29676035046577454	0.2736732570239334	0.8427456149244652	0.610255370235793	0.001
26	0.2955995500087738	0.2754942767950052	0.8420542140997499	0.6045462014226007	0.001
27	0.29585039615631104	0.27653485952133194	0.8437684356323902	0.6115221375683754	0.001
28	0.295540988445282	0.26925078043704476	0.8446938104986479	0.6191186747828321	0.001
29	0.3010655343532562	0.2663891779396462	0.8437664387164651	0.6215750043898619	0.001
30	0.29214760661125183	0.2809573361082206	0.8437435686355217	0.6025311078598518	0.0001
31	0.29040178656578064	0.28121748178980227	0.8439103638567266	0.6071651131848005	0.0001
32	0.29034462571144104	0.2809573361082206	0.8437194965322373	0.6111569473926136	0.0001
33	0.2888760268688202	0.28537981269510926	0.8461617038663874	0.6202495870793918	0.0001
34	0.28964364528656006	0.2861602497398543	0.8446023671361742	0.6150504150317478	0.0001
35	0.28874215483665466	0.2866805411030177	0.8449244728566273	0.611180048847438	0.0001
36	0.2888963222503662	0.28355879292403746	0.8447173058645225	0.6119874534823754	0.0001
37	0.288282573223114	0.2866805411030177	0.8475834540970686	0.6255767175486281	0.0001
38	0.29050976037979126	0.28251821019771073	0.8452536426724028	0.6057239934398935	0.0001
39	0.28778275847435	0.28537981269510926	0.8470600182796791	0.625366961909805	0.0001
40	0.2885717749595642	0.2809573361082206	0.8468000302716884	0.622337777946806	0.0001
41	0.28773826360702515	0.2843392299687825	0.847323400258903	0.6260539681026288	0.0001
42	0.28776827454566956	0.28563995837669093	0.8476613005450627	0.6199392946357273	0.0001
43	0.28717148303985596	0.28303850156087407	0.8479237095716232	0.6287571427217789	0.0001
44	0.28678667545318604	0.28407908428720086	0.8463665693654939	0.6189979239207937	0.0001
45	0.28698909282684326	0.28381893860561913	0.8462928555066304	0.6235508782461164	0.0001
46	0.2868472635746002	0.28251821019771073	0.8459846547314578	0.6151318511304835	0.0001
47	0.28715068101882935	0.2845993756503642	0.8462129359348595	0.6211457155619424	0.0001
48	0.28661593794822693	0.28355879292403746	0.8466852933705867	0.6231150403485404	0.0001
49	0.28633347153663635	0.28590010405827265	0.8460415439387342	0.616055362439494	0.0001
50	0.28642749786376953	0.2845993756503642	0.8482882700250868	0.625458075101288	0.0001
51	0.2890762686729431	0.28485952133194586	0.848592785832539	0.6278100779578839	0.0001
52	0.2855978012084961	0.2851196670135276	0.8464228285561143	0.6255462096645672	0.0001
53	0.2872205674648285	0.27887617065556713	0.8489991514001897	0.6457587856102145	0.0001
54	0.2855803072452545	0.2903225806451613	0.8476844874709444	0.6243869856844756	0.0001
55	0.28568968176841736	0.2845993756503642	0.8475136716266056	0.6339630509281279	0.0001
56	0.28617897629737854	0.2866805411030177	0.8465597622829039	0.6241465491773776	0.0001
57	0.2870914936065674	0.2861602497398543	0.845436853426201	0.6249269702519318	0.0001
58	0.2857914865016937	0.28121748178980227	0.8491941382702348	0.6333866717026029	0.0001
59	0.28617140650749207	0.2887617065556712	0.8468232576049287	0.6178461796051926	1e-05
60	0.2846605181694031	0.28537981269510926	0.8485033598045205	0.6275748058546806	1e-05
61	0.2848633825778961	0.28303850156087407	0.8479865171982329	0.6223888517425455	1e-05
62	0.28548601269721985	0.2843392299687825	0.8469200122586577	0.6247632003821695	1e-05
63	0.28493326902389526	0.2827783558792924	0.8488979777323336	0.6274806463168713	1e-05
64	0.28459736704826355	0.28225806451612906	0.8475187206498287	0.6370787064578803	1e-05
65	0.2860054671764374	0.2869406867845994	0.8467700785794469	0.6240984315849201	1e-05
66	0.2847185730934143	0.28407908428720086	0.8481340441736481	0.6346693986906206	1e-05
67	0.28529325127601624	0.28537981269510926	0.8487528745798691	0.6287121285420982	1e-05
68	0.2852926254272461	0.2866805411030177	0.8480251642525557	0.6321379394582358	1e-05
69	0.284834623336792	0.28355879292403746	0.847692190707931	0.6397237492354447	1e-05
70	0.28527727723121643	0.28225806451612906	0.8492167101827677	0.6381143671040704	1e-05
71	0.28507113456726074	0.2882414151925078	0.8475971370143149	0.6325489300082728	1.0000000000000002e-06
72	0.28452861309051514	0.28485952133194586	0.8474255781269963	0.6236352127811986	1.0000000000000002e-06
73	0.28448227047920227	0.28121748178980227	0.847641772858811	0.6333277250193455	1.0000000000000002e-06
74	0.28447526693344116	0.2827783558792924	0.8465770953294945	0.6300187593616763	1.0000000000000002e-06
75	0.2851284146308899	0.28199791883454733	0.8473772748126625	0.6235297745568456	1.0000000000000002e-06
76	0.2859683036804199	0.2879812695109261	0.847320835674516	0.6186062513830065	1.0000000000000002e-06
77	0.2858298718929291	0.28563995837669093	0.8459046737621472	0.6172786558676017	1.0000000000000002e-06
78	0.28438833355903625	0.2843392299687825	0.8480547459130655	0.6325947858436887	1.0000000000000002e-06
79	0.2870919704437256	0.2874609781477627	0.8472353346431579	0.617917490234713	1.0000000000000002e-06
80	0.28482332825660706	0.28381893860561913	0.8477330616403465	0.6286567457369128	1.0000000000000002e-06
81	0.2847617268562317	0.28537981269510926	0.8489678202792957	0.6304525529970205	1.0000000000000002e-06
82	0.28511229157447815	0.28590010405827265	0.8480416961845967	0.6394217270135759	1.0000000000000002e-06
83	0.284644216299057	0.28563995837669093	0.8488055562622434	0.6255055774993536	1.0000000000000002e-06
84	0.2857225835323334	0.2832986472424558	0.848188643119867	0.6457553263622914	1.0000000000000002e-06
85	0.28550758957862854	0.28121748178980227	0.848818698673405	0.6339586571635658	1.0000000000000002e-07
86	0.284895658493042	0.28590010405827265	0.8479890588592848	0.6362631688004041	1.0000000000000002e-07
87	0.2845035493373871	0.2851196670135276	0.8473590201582036	0.6327749126527296	1.0000000000000002e-07
88	0.28541097044944763	0.28121748178980227	0.8477551536613127	0.6370893160624239	1.0000000000000002e-07

Framework Versions

Transformers: 4.41.0
Pytorch: 2.5.0+cu124
Datasets: 3.0.2
Tokenizers: 0.19.1

lombardata
/

drone-DinoVdeau-produttoria_binary-binary-large-2024_11_03-batch-size64_freeze