Storage - where the model weights are stored: Supervisely Server or some Node from your Cluster
Managing your models
Click the "three dots" icon to see all available actions for the model.
1.
Goto task - redirects to the task page and filters all tasks. User will be able to get all necessary information about training: logs, charts, checkpoints, etc.
2.
Edit - change the model name
3.
Upload to Supervisely - if the model is stored on some Node from Cluster, user can upload this model to the Supervisely Server. That allows using this model on other computational Nodes. Read mode on NN lifecycle
4.
Archive
5.
Config - training configuration (it is useful to see all training parameters including classes the model predicts).
Here is the example of the entire training configuration for our custom YOLO V3.
1
{
2
"settings":{
3
"lr":0.0001,
4
"epochs":15,
5
"batch_size":{
6
"train":12
7
},
8
"input_size":{
9
"width":256,
10
"height":256
11
},
12
"bn_momentum":0.07,
13
"gpu_devices":[
14
0,
15
1,
16
2,
17
3
18
],
19
"data_workers":{
20
"train":3
21
},
22
"dataset_tags":{
23
"train":"train"
24
},
25
"subdivisions":{
26
"train":1
27
},
28
"print_every_iter":10,
29
"weights_init_type":"transfer_learning",
30
"enable_augmentations":false
31
},
32
"out_classes":[
33
{
34
"color":"#82789c",
35
"shape":"rectangle",
36
"title":"kiwi_bbox"
37
},
38
{
39
"color":"#57c0cb",
40
"shape":"rectangle",
41
"title":"lemon_bbox"
42
}
43
],
44
"class_title_to_idx":{
45
"kiwi_bbox":0,
46
"lemon_bbox":1
47
}
48
}
Copied!
settings - training parameters
out_classes - classes model predicts
class_title_to_idx - mapping from class title to class index
Downloading models
If the model is stored on some Node from you Cluster, you have to upload it to the Supervisely Server (manage models). After that the "Download" button will be available. As a result user will download .tar archive with model weights.
Starting training / inference
From this page you can start training or inference. Please read corresponding chapters to get more information.