groderg/SegIGNCoral-b0-2025_10_09_40484-bs16_refine

SegIGNCoral-b0-2025_10_09_40484-bs16_refine is a fine-tuned version of nvidia/mit-b0.

---

Model description

SegIGNCoral-b0-2025_10_09_40484-bs16_refine is a model built on top of nvidia/mit-b0 model for aerial image segmentation.

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 procedure

Training hyperparameters

The following hyperparameters were used during training:

• Number of Epochs: 175.0
• Learning Rate: 1e-05
• Train Batch Size: 16
• Eval Batch Size: 16
• Optimizer: Adam
• LR Scheduler Type: ReduceLROnPlateau with a patience of 5 epochs and a factor of 0.1
• Freeze Encoder: No
• Data Augmentation: No

Training results

Evaluate on training set.

{
    "SegIGNCoral-b0-2025_10_09_40484-bs16_refine": {
        "pixel_acc": 0.8066,
        "mean_iou": 0.5697,
        "iou_per_class": {
            "Acropora Branching": 0.6439,
            "Non-acropora Massive": 0.2779,
            "Other Corals": 0.3327,
            "Sand": 0.821,
            "Seagrass": 0.7729
        }
    }
}

Evaluate on manually annotate zone.

📊 Evaluating zone: hermitage
  ✅ Pixel Accuracy: 0.8814
  ✅ Mean Accuracy : 0.3494
  ✅ Mean IoU     : 0.3283
  Pixel Accuracy Per Class:
    Acropora Branching: 0.0000
    Non-acropora Massive: 0.0000
    Other Corals: 0.0000
    Sand: 0.9011
    Syringodium: 0.8457
  IoU Per Class:
    Acropora Branching: 0.0000
    Non-acropora Massive: 0.0000
    Other Corals: 0.0000
    Sand: 0.8805
    Syringodium: 0.7611

📊 Evaluating zone: troudeau
  ✅ Pixel Accuracy: 0.8389
  ✅ Mean Accuracy : 0.5461
  ✅ Mean IoU     : 0.3493
  Pixel Accuracy Per Class:
    Acropora Branching: 0.5263
    Non-acropora Massive: 0.3429
    Other Corals: 0.4113
    Sand: 0.9040
  IoU Per Class:
    Acropora Branching: 0.0481
    Non-acropora Massive: 0.1714
    Other Corals: 0.2827
    Sand: 0.8949

📊 Evaluating zone: stleu
  ✅ Pixel Accuracy: 0.6907
  ✅ Mean Accuracy : 0.6036
  ✅ Mean IoU     : 0.3995
  Pixel Accuracy Per Class:
    Acropora Branching: 0.9457
    Non-acropora Massive: 0.5060
    Other Corals: 0.1624
    Sand: 0.8006
  IoU Per Class:
    Acropora Branching: 0.3820
    Non-acropora Massive: 0.3172
    Other Corals: 0.1358
    Sand: 0.7631

📦 Micro-Averaged Metrics Across Zones (all pixels):
  Pixel Accuracy Per Class:
    Acropora Branching: 0.9351
    Non-acropora Massive: 0.4778
    Other Corals: 0.2117
    Sand: 0.8996
    Syringodium: 0.8457
  IoU Per Class:
    Acropora Branching: 0.0906
    Non-acropora Massive: 0.1418
    Other Corals: 0.0172
    Sand: 0.8788
    Syringodium: 0.7611

  ✅ Pixel Accuracy: 0.8781
  ✅ Mean Accuracy : 0.6740
  ✅ Mean IoU     : 0.3779

Epoch	Validation Loss	Learning Rate
1.0	0.17825450003147125	1e-05
2.0	0.17652319371700287	1e-05
3.0	0.17437469959259033	1e-05
4.0	0.17109064757823944	1e-05
5.0	0.16746488213539124	1e-05
6.0	0.1647380143404007	1e-05
7.0	0.1629670113325119	1e-05
8.0	0.16162735223770142	1e-05
9.0	0.16086645424365997	1e-05
10.0	0.1595691293478012	1e-05
11.0	0.1584334373474121	1e-05
12.0	0.15793322026729584	1e-05
13.0	0.15778672695159912	1e-05
14.0	0.1567489206790924	1e-05
15.0	0.15607595443725586	1e-05
16.0	0.15554767847061157	1e-05
17.0	0.1547158807516098	1e-05
18.0	0.1543949544429779	1e-05
19.0	0.15392234921455383	1e-05
20.0	0.15279631316661835	1e-05
21.0	0.15245354175567627	1e-05
22.0	0.152305468916893	1e-05
23.0	0.15164558589458466	1e-05
24.0	0.15041886270046234	1e-05
25.0	0.1490705907344818	1e-05
26.0	0.14869147539138794	1e-05
27.0	0.14790189266204834	1e-05
28.0	0.146309033036232	1e-05
29.0	0.14488185942173004	1e-05
30.0	0.14381851255893707	1e-05
31.0	0.14236079156398773	1e-05
32.0	0.14145053923130035	1e-05
33.0	0.14065156877040863	1e-05
34.0	0.14140239357948303	1e-05
35.0	0.13949288427829742	1e-05
36.0	0.13923388719558716	1e-05
37.0	0.1388508528470993	1e-05
38.0	0.1380811184644699	1e-05
39.0	0.13670456409454346	1e-05
40.0	0.13674621284008026	1e-05
41.0	0.13670817017555237	1e-05
42.0	0.13663434982299805	1e-05
43.0	0.1345621645450592	1e-05
44.0	0.1353185921907425	1e-05
45.0	0.13393084704875946	1e-05
46.0	0.13347546756267548	1e-05
47.0	0.13446684181690216	1e-05
48.0	0.13390202820301056	1e-05
49.0	0.1328386515378952	1e-05
50.0	0.13314343988895416	1e-05
51.0	0.13273906707763672	1e-05
52.0	0.13288834691047668	1e-05
53.0	0.13220956921577454	1e-05
54.0	0.13126344978809357	1e-05
55.0	0.131490096449852	1e-05
56.0	0.13168570399284363	1e-05
57.0	0.13094131648540497	1e-05
58.0	0.13041794300079346	1e-05
59.0	0.12996184825897217	1e-05
60.0	0.13028591871261597	1e-05
61.0	0.1296725869178772	1e-05
62.0	0.1295732855796814	1e-05
63.0	0.1319553703069687	1e-05
64.0	0.12922300398349762	1e-05
65.0	0.12969540059566498	1e-05
66.0	0.1294255256652832	1e-05
67.0	0.12892434000968933	1e-05
68.0	0.13009504973888397	1e-05
69.0	0.1286858767271042	1e-05
70.0	0.12956416606903076	1e-05
71.0	0.12911681830883026	1e-05
72.0	0.1296709179878235	1e-05
73.0	0.1287732869386673	1e-05
74.0	0.1294514536857605	1e-05
75.0	0.12915697693824768	1e-05
76.0	0.12841865420341492	1e-05
77.0	0.12817372381687164	1e-05
78.0	0.1281365156173706	1e-05
79.0	0.1279040277004242	1e-05
80.0	0.1278274655342102	1e-05
81.0	0.1280566155910492	1e-05
82.0	0.12727053463459015	1e-05
83.0	0.12869715690612793	1e-05
84.0	0.12711170315742493	1e-05
85.0	0.12702301144599915	1e-05
86.0	0.12730672955513	1e-05
87.0	0.12725737690925598	1e-05
88.0	0.1271396279335022	1e-05
89.0	0.12912477552890778	1e-05
90.0	0.12642411887645721	1e-05
91.0	0.12625634670257568	1e-05
92.0	0.127320796251297	1e-05
93.0	0.1259581744670868	1e-05
94.0	0.1271350383758545	1e-05
95.0	0.12815390527248383	1e-05
96.0	0.1261141449213028	1e-05
97.0	0.12658798694610596	1e-05
98.0	0.1271129697561264	1e-05
99.0	0.12775172293186188	1e-05
100.0	0.12581050395965576	1e-05
101.0	0.12726755440235138	1e-05
102.0	0.12562812864780426	1e-05
103.0	0.12696777284145355	1e-05
104.0	0.12738262116909027	1e-05
105.0	0.12724792957305908	1e-05
106.0	0.12833286821842194	1e-05
107.0	0.12599430978298187	1e-05
108.0	0.12838247418403625	1e-05
109.0	0.12654130160808563	1e-05
110.0	0.12597838044166565	1e-05
111.0	0.12549486756324768	1e-05
112.0	0.12560401856899261	1e-05
113.0	0.1261509358882904	1e-05
114.0	0.12717725336551666	1e-05
115.0	0.1258397251367569	1e-05
116.0	0.12457232922315598	1e-05
117.0	0.125536248087883	1e-05
118.0	0.12458962202072144	1e-05
119.0	0.1279805451631546	1e-05
120.0	0.12609897553920746	1e-05
121.0	0.12450611591339111	1e-05
122.0	0.1254856139421463	1e-05
123.0	0.12441741675138474	1e-05
124.0	0.1260562390089035	1e-05
125.0	0.12651772797107697	1e-05
126.0	0.12526915967464447	1e-05
127.0	0.12434830516576767	1e-05
128.0	0.1253136843442917	1e-05
129.0	0.12462199479341507	1e-05
130.0	0.12565144896507263	1e-05
131.0	0.12453299760818481	1e-05
132.0	0.1274179220199585	1e-05
133.0	0.1271260529756546	1e-05
134.0	0.12498216331005096	1e-05
135.0	0.1252247840166092	1e-05
136.0	0.12617097795009613	1e-05
137.0	0.12494047731161118	1e-05
138.0	0.1247497946023941	1e-05
139.0	0.1245729923248291	1.0000000000000002e-06
140.0	0.12425211817026138	1.0000000000000002e-06
141.0	0.12417663633823395	1.0000000000000002e-06
142.0	0.12412694841623306	1.0000000000000002e-06
143.0	0.12370014935731888	1.0000000000000002e-06
144.0	0.12406766414642334	1.0000000000000002e-06
145.0	0.12380032241344452	1.0000000000000002e-06
146.0	0.12423602491617203	1.0000000000000002e-06
147.0	0.12404759973287582	1.0000000000000002e-06
148.0	0.12442447990179062	1.0000000000000002e-06
149.0	0.12393788993358612	1.0000000000000002e-06
150.0	0.12402787059545517	1.0000000000000002e-06
151.0	0.12422869354486465	1.0000000000000002e-06
152.0	0.12365838885307312	1.0000000000000002e-06
153.0	0.12383443862199783	1.0000000000000002e-06
154.0	0.12431268393993378	1.0000000000000002e-06
155.0	0.12341240048408508	1.0000000000000002e-06
156.0	0.12390484660863876	1.0000000000000002e-06
157.0	0.12473750114440918	1.0000000000000002e-06
158.0	0.12396185100078583	1.0000000000000002e-06
159.0	0.12398138642311096	1.0000000000000002e-06
160.0	0.1234150305390358	1.0000000000000002e-06
161.0	0.12368131428956985	1.0000000000000002e-06
162.0	0.12391027063131332	1.0000000000000002e-06
163.0	0.12366586923599243	1.0000000000000002e-06
164.0	0.1238330602645874	1.0000000000000002e-06
165.0	0.12347564101219177	1.0000000000000002e-06
166.0	0.12349989265203476	1.0000000000000002e-06
167.0	0.12391113489866257	1.0000000000000002e-07
168.0	0.12376048415899277	1.0000000000000002e-07
169.0	0.12376976013183594	1.0000000000000002e-07
170.0	0.12379935383796692	1.0000000000000002e-07
171.0	0.12354343384504318	1.0000000000000002e-07
172.0	0.12349338084459305	1.0000000000000002e-07
173.0	0.12366225570440292	1.0000000000000002e-07
174.0	0.12352761626243591	1.0000000000000002e-07
175.0	0.12430573999881744	1.0000000000000002e-07

---

Framework Versions

• Transformers: 4.56.2
• Pytorch: 2.8.0+cu128
• Datasets: 4.1.1
• Tokenizers: 0.22.1