-
Notifications
You must be signed in to change notification settings - Fork 0
/
flower-classification-results.txt
73 lines (53 loc) · 2.69 KB
/
flower-classification-results.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
fastai
Resnet32, 5+5 epochs, 64 bs, lr=0.02
epoch train_loss valid_loss accuracy
5 0.059245 0.104859 0.977995
Resnet152, 5+5 epochs, 32 bs, lr=0.01
epoch train_loss valid_loss accuracy
5 0.036988 0.105719 0.986553
Densenet121, 5+5 epochs, 32 bs, lr=0.02
epoch train_loss valid_loss accuracy
5 0.046098 0.097080 0.981663
pytorch
0.25 subset
resnet34, 15 epochs, lr=0.01/0.001 StepRL schedule 5, basic affine transforms, 45 deg rot
Best val Loss: 0.429578, Best val Acc: 0.891198
Best val Loss: 0.354743, Best val Acc: 0.922983
full dataset
resnet34, 15 epochs
Best val Acc: 0.863081
Best val Acc: 0.900978
resnet34, 15 epochs, lr=0.01/0.001 StepRL schedule 5, basic affine transforms, 45 deg rot
Best val Loss: 0.248217, Best val Acc: 0.941320
Best val Loss: 0.131558, Best val Acc: 0.974328
resnet34, 15 epochs, 1cycle + other tricks from fastai, basic affine transforms, 45 deg rot
Best val Loss: 0.246839, Best val Acc: 0.943765
Best val Loss: 0.093415, Best val Acc: 0.982885
resnet34, 15 epochs, 1cycle + other tricks from fastai, AdamW wd=0.01/0.1, basic affine transforms, 45 deg rot
Best val Loss: 0.230131, Best val Acc: 0.948655
Best val Loss: 0.078676, Best val Acc: 0.987775
resnet34, 15 epochs, 1cycle lr=1e-2/3e-4 + other tricks from fastai, AdamW wd=0.01/0.01, basic affine transforms, 45 deg rot, 10xTTA
Best val Loss: 0.138446, Best val Acc: 0.965770
Best val Loss: 0.068852, Best val Acc: 0.987775
resnet101, 15 epochs, bs 48/64, 1cycle lr=3e-3/7e-5 + other tricks from fastai, AdamW wd=5e-2/2e-4, basic affine transforms, 45 deg rot, 10xTTA
Best val Loss: 0.111695, Best val Acc: 0.968215
Best val Loss: 0.054569, Best val Acc: 0.992665
resnet152, 15 epochs, bs 48/64, 1cycle lr=3e-3/7e-5 + other tricks from fastai, AdamW wd=5e-2/2e-4, basic affine transforms, 45 deg rot, 10xTTA
Best val Loss: 0.114294, Best val Acc: 0.976773
Best val Loss: 0.069883, Best val Acc: 0.991443
resnet34, 15 epochs, pretrain on 128, 1cycle lr=1e-2/3e-4/1e-2/1e-4 + other tricks from fastai, AdamW wd=0.01/0.01/0.01/0.1, basic affine transforms, 45 deg rot, 10xTTA
Best val Loss: 0.193959, Best val Acc: 0.943765
Best val Loss: 0.084297, Best val Acc: 0.982885
Best val Loss: 0.093187, Best val Acc: 0.988998
Best val Loss: 0.098492, Best val Acc: 0.990220
5CV
resnet50, 15 epochs, 1cycle lr=1e-2/1e-4 + other tricks from fastai, AdamW wd=0.01/0.1, basic affine transforms, 45 deg rot, 10xTTA
mean_acc 0.987042
mean_acc_head 0.964303
mean_loss 0.082470
mean_loss_head 0.147141
resnet101, 15 epochs, bs 48/64, 1cycle + other tricks from fastai, lr=3e-3/7e-5, AdamW wd=5e-2/2e-4, basic affine transforms, 45 deg rot, 10xTTA
mean_acc 0.988467
mean_acc_head 0.969335
mean_loss 0.059509
mean_loss_head 0.127831