reports/training_final_report.md

SID-UNet Evaluation Report

Run Configuration & Hyperparameters

Hyperparameter / Setting	Configured Value
Project Name	test_smoke
Dataset	KhangTruong/IMD2020
Splits	Train: 'train'
Streaming	False
Image Size	[32, 32]
Batch Size	16
Epoch Limit / Steps	2 samples/epoch
Model	unet (features=[8, 16], bilinear=True)
Aux Classifier	False (classes=3)
Mask Loss	Combined (BCE: 0.5, Dice: 0.5, Focal: 0.5)
Aux Loss	cross_entropy (wt=0.2)
Post-Processing	Enabled (min_area=64, fill_holes=True, morph=open_close)
Optimizer	adamw (lr=0.001, wd=0.0001)
Scheduler	cosine
Epochs	1
AMP	False

Overall Segmentation & Classification Metrics

Metric	Value
Val Bce Loss	0.6957
Val Dice Loss	0.608
Val Focal Loss	0.0759
Val Mask Loss	0.6898
Val Total Loss	0.6898
Val Iou	0.1676
Val Dice	0.2576
Val F1	0.2576
Val Pixel F1	0.2576
Val Auroc	0.69
Val Pixel Auroc	0.69
Val Pixel Acc	0.485
Val Precision	0.6
Val Recall	0.1676
Val Specificity	0.9174

Training & Validation Curves

Epoch-by-Epoch Training & Validation Progression

Epoch	Train Loss	Val Loss	Val IoU	Val F1	Val AUROC	Val Pixel Acc	Learning Rate
1	0.6793	0.6806	0.6	0.6	0.7947	0.6	0.001
2	0.8813	0.9076	0.0129	0.0254	0.5057	0.0143	0.000251
3	0.6155	0.6859	0.1782	0.2147	0.6987	0.416	0.000147
4	0.7411	0.6793	0.6	0.6	0.7932	0.6	0.001
5	0.6793	0.6816	0.6	0.6	0.7959	0.6	6.79e-05
6	0.6755	0.6836	0.5813	0.5904	0.7982	0.591	5.05e-05
7	0.6724	0.6858	0.0902	0.1493	0.6492	0.3758	3.9e-05
8	0.67	0.6878	0.0186	0.036	0.6962	0.4041	3.11e-05
9	0.668	0.6899	0.015	0.0292	0.6933	0.4035	2.54e-05
10	0.6664	0.6917	0.0142	0.0275	0.6906	0.4023	2.12e-05
11	0.6651	0.6929	0.0151	0.0294	0.6882	0.403	1.8e-05
12	0.664	0.6951	0.0158	0.0307	0.6857	0.4033	1.55e-05
13	0.663	0.6981	0.0176	0.0341	0.6831	0.3997	1.35e-05
14	0.6621	0.7005	0.0287	0.0545	0.6809	0.4055	1.19e-05
15	0.6614	0.7018	0.0394	0.0728	0.6801	0.4048	1.06e-05
16	0.6607	0.6998	0.0573	0.1017	0.6816	0.4112	9.5e-06
17	0.6601	0.6964	0.0835	0.1404	0.6843	0.423	8.59e-06
18	0.6596	0.6927	0.1124	0.1821	0.687	0.4475	7.81e-06
19	0.6591	0.69	0.1313	0.2095	0.6889	0.4612	7.15e-06
20	0.6586	0.6888	0.1456	0.2294	0.6899	0.4709	6.58e-06
21	0.6582	0.6893	0.1593	0.2474	0.69	0.479	6.08e-06
22	0.6578	0.6898	0.1676	0.2576	0.69	0.485	1e-06

Per-Subset Metrics Breakdown

Subset	Mean IoU	Dice / F1	Pixel Accuracy	Sample Count
Label 0 (Real / Black Mask)	0	0	0.7935	4
Label 1 (Synthetic / White Mask)	0.3099	0.4658	0.3099	3
Label 2 (Tampered / Partial Mask)	0.2487	0.3929	0.2487	3