-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconfig.yaml
91 lines (86 loc) · 2.58 KB
/
config.yaml
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
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
epochs: 30 # number of epochs to train
num_class: 10
data_dir: 'resources/numta'
output_dir: 'output'
train_batch: 128
valid_batch: 128
lr: 0.01
debug: false
model: # model3 configuration
- [Conv2D, [3, 16, 3, 1, 1]] # in_channels, out_channels, kernel_size, stride=1, padding=0, bias=True
- [ReLU, []]
- [MaxPool2D, [2, 2]]
- [Conv2D, [16, 32, 3, 1, 1]]
- [ReLU, []]
- [MaxPool2D, [2, 2]]
- [Conv2D, [32, 32, 3, 1, 1]]
- [ReLU, []]
- [MaxPool2D, [2, 2]] # kernel_size, stride
- [Flatten, []]
- [Linear, [1024]]
- [ReLU, []]
- [Linear, [512]]
- [ReLU, []]
- [Linear, [10]]
- [Softmax, []]
checkpoint_path: "resources/1705044_model.npy" # checkpoint to load weights from
resume: false # whether to continue training from checkpoint
gt_csv: false #"resources/numta/training-d.csv" # ground truth csv for test.py
augment:
img_shape: [48, 48]
use_bbox: true # crop to bbox
reverse: true # inverse image
aug: false # use rotate, blur
dilation: true # use dilation op
opening: false # use opening op
mixup: 0.0 # probability of using mixup
contour_cutout_prob: 0.5 # probability of using contour cutout
contour_cutout_number: 5 # highest number of contour cutouts from an image
cache: true # whether to cache the dataset first -> saves time in training
lr_scheduler:
factor: 0.7
patience: 2
use_wandb: false
wandb:
project: "cse472_cnn_scratch"
entity: "nexh98" # anonymous
notebook: ''
name: 'Baseline'
comment: 'Model-48x48'
exp: 'Baseline'
# model2 configuration
# model
# - [Conv2D, [3, 6, 5, 1, 0]] # in_channels, out_channels, kernel_size, stride=1, padding=0, bias=True
# - [ReLU, []]
# - [MaxPool2D, [2, 2]]
# - [Conv2D, [6, 16, 5, 1, 0]]
# - [ReLU, []]
# - [MaxPool2D, [2, 2]]
# - [Conv2D, [16, 32, 3, 1, 1]]
# - [ReLU, []]
# - [MaxPool2D, [2, 2]]
# - [Flatten, []] # 64 * 5 * 5
# - [Linear, [120]] #
# - [ReLU, []] # 26
# - [Linear, [84]] #
# - [ReLU, []] # 26
# - [Linear, [10]] #
# - [Softmax, []]
# model3
# model:
# - [Conv2D, [3, 16, 3, 1, 1]] # in_channels, out_channels, kernel_size, stride=1, padding=0, bias=True
# - [ReLU, []]
# - [MaxPool2D, [2, 2]]
# - [Conv2D, [16, 32, 3, 1, 1]]
# - [ReLU, []]
# - [MaxPool2D, [2, 2]]
# - [Conv2D, [32, 32, 3, 1, 1]]
# - [ReLU, []] # 26
# - [MaxPool2D, [2, 2]] # kernel_size, stride=1 -> 13
# - [Flatten, []] # 64 * 5 * 5
# - [Linear, [1024]] #
# - [ReLU, []] # 26
# - [Linear, [512]] #
# - [ReLU, []] # 26
# - [Linear, [10]] #
# - [Softmax, []]