hourouu commited on
Commit
53450aa
Β·
verified Β·
1 Parent(s): de4dfc7

Results: results_report.txt

Browse files
Files changed (1) hide show
  1. results2/results_report.txt +55 -0
results2/results_report.txt ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ =================================================================
2
+ UNet v17 β€” Lung Segmentation β€” Results Report
3
+ =================================================================
4
+
5
+ Total epochs trained : 25
6
+ Best epoch : 12
7
+
8
+ ─────────────────────────────────────────────────────────────────
9
+ BEST VALIDATION CHECKPOINT (Epoch 12)
10
+ ─────────────────────────────────────────────────────────────────
11
+ Dice (overall) : 0.9047
12
+ Dice (lung only) : 0.8787
13
+ IoU : 0.8718
14
+ Sensitivity : 0.9427
15
+ Specificity : 0.9879
16
+ Val Loss : 0.0978
17
+ Train Dice : 0.9189
18
+ Train Loss : 0.0267
19
+ LR at best epoch : 0.000021
20
+
21
+ ─────────────────────────────────────────────────────────────────
22
+ LAST EPOCH (25)
23
+ ─────────────────────────────────────────────────────────────────
24
+ Dice (overall) : 0.9026
25
+ Dice (lung only) : 0.8767
26
+ IoU : 0.8695
27
+ Sensitivity : 0.9396
28
+ Specificity : 0.9878
29
+ Val Loss : 0.1078
30
+
31
+ ─────────────────────────────────────────────────────────────────
32
+ HYPERPARAMETERS (from Cell 4)
33
+ ─────────────────────────────────────────────────────────────────
34
+ Batch size : 8
35
+ LR (initial) : 1e-4
36
+ Optimizer : AdamW (weight_decay=1e-4)
37
+ Scheduler : LinearWarmup(5ep) β†’ CosineWarmRestarts
38
+ Loss : 0.5Β·BCE + 0.5Β·Dice
39
+ Dropout enc shallow : 0.10
40
+ Dropout enc deep : 0.20
41
+ Dropout bottleneck : 0.30
42
+ Dropout decoder : 0.10
43
+ Max epochs : 70
44
+ Early stopping patience: 20
45
+
46
+ ─────────────────────────────────────────────────────────────────
47
+ MODEL
48
+ ─────────────────────────────────────────────────────────────────
49
+ Architecture : UNetV3 (2-D U-Net + spatial Dropout2d)
50
+ Base channels : 32
51
+ Normalization : InstanceNorm2d (affine=True)
52
+ Input : 1-channel CT slice
53
+ Task : Binary lung segmentation
54
+
55
+ =================================================================