📌
Getting started
📂
Data Organization
📝
Labeling
🤝
Collaboration
⚙
Data Manipulation
🔮
Neural Networks
👔
Enterprise Edition
If
"if" layer (
if) is used to split input data to several flows with a specified criterion. Lets consider all possible use cases.Use case: split with probability
1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"probability": 0.95
11
}
12
}
13
}
Copied!
In this case input data (image + annotation) will go to the "left" branch (
$data_1 varibable) with probability 95%, to the "right" branch ($data_2 varibable) with probability 5%.It can be used in several scenarios.
Split data to train and val subsets
First scenario is when we are going to split data into train and validation sets by adding corresponding tag to each image.
1
[
2
{
3
"action": "data",
4
"src": ["my_proj/*"],
5
"dst": "$data",
6
"settings": {
7
"classes_mapping": "default"
8
}
9
},
10
{
11
"action": "if",
12
"src": ["$data"],
13
"dst": [
14
"$totrain",
15
"$toval"
16
],
17
"settings": {
18
"condition": {
19
"probability": 0.95
20
}
21
}
22
},
23
{
24
"action": "tag",
25
"src": ["$totrain"],
26
"dst": "$train",
27
"settings": {
28
"tag": "train",
29
"action": "add"
30
}
31
},
32
{
33
"action": "tag",
34
"src": ["$toval"],
35
"dst": "$val",
36
"settings": {
37
"tag": "val",
38
"action": "add"
39
}
40
},
41
{
42
"action": "supervisely",
43
"src": [
44
"$train",
45
"$val"
46
],
47
"dst": "proj_with_tags",
48
"settings": {}
49
}
50
]
Copied!
Computational graph:
Get sample from data
Second scenario is when we are going to get small data sample:
1
[
2
{
3
"action": "data",
4
"src": ["my_proj/*"],
5
"dst": "$data",
6
"settings": {
7
"classes_mapping": "default"
8
}
9
},
10
{
11
"action": "if",
12
"src": ["$data"],
13
"dst": [
14
"$data_sample",
15
"$data_ignore"
16
],
17
"settings": {
18
"condition": {
19
"probability": 0.01
20
}
21
}
22
},
23
{
24
"action": "supervisely",
25
"src": ["$data_sample"],
26
"dst": "proj_sample",
27
"settings": {}
28
}
29
]
Copied!
Computational graph:
We just get 1 persent of data and put it to project
proj_sample. Other data will be skipped.Use case: split data by objects count
This case allows us to split data by objects count (
min_objects_count value). In the example below, images with equal or more that three objects will be passed to $data_1 branch, other images — to the $data_2 branch1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"min_objects_count": 3
11
}
12
}
13
}
Copied!
Example of full config:
1
[
2
{
3
"action": "data",
4
"src": ["my_proj/*"],
5
"dst": "$data",
6
"settings": {
7
"classes_mapping": "default"
8
}
9
},
10
{
11
"action": "if",
12
"src": ["$data" ],
13
"dst": [
14
"$data_1",
15
"$data_2"
16
],
17
"settings": {
18
"condition": {
19
"min_objects_count": 3
20
}
21
}
22
},
23
{
24
"action": "supervisely",
25
"src": ["$data_1"],
26
"dst": "proj_filter",
27
"settings": {}
28
}
29
]
Copied!
Computational graph:
In this example we put images with
"min_objects_count": 3 to proj_filter. Other images are ignored.Use case: split by image height or width
For example, we are going to filter images by minimum image height (200 pixels):
1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"min_height": 200
11
}
12
}
13
}
Copied!
The same can be applied to minimum width:
1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"min_width": 200
11
}
12
}
13
}
Copied!
This case is useful when we are going to skip some small images after applying "instance crop" layer.
Use case: split by tags
This example passes images with tag "party" or "dinner" to the
$data_1 branch, other images are passed to $data_2 branch.1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"tags": ["party", "dinner"]
11
}
12
}
13
}
Copied!
Use case: split by class presence
In this example images which contain any object of some required class (
person or dog) are passed to the $data_1 branch.1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"include_classes": ["person", "dog"]
11
}
12
}
13
}
Copied!
Use case: split by images names in range, and step
Names is ordered alphabetical (a, b ,c ...).
1
{
2
"action": "if",
3
"src": ["$data_input" ],
4
"dst": [
5
"$data_1",
6
"$data_2"
7
],
8
"settings": {
9
"condition": {
10
"name_in_range": ["a0100", "a0105"],
11
"frame_step": 2
12
}
13
}
14
}
Copied!
Should pass images with names ["00100", "00102", "00104"] into first branch (
$data1) and else into ($data2) branch.Last modified 1yr ago