CustomModelsSelector

Introduction

CustomModelsSelector widget allows creating a table with NN models that you have trained in Supervisely by providing path to training application default save directory, Team ID and task type (e.g. object detection, instance segmentation, pose estimation and etc).

Read this tutorial in developer portal.

Function signature

CustomModelsSelector(
    team_id = team_id,
    training_app_directory="/yolov8_train/",
    task_type="object_detection",
    widget_id=None,
)

custom-models-selector

Parameters

Parameters	Type	Description
team_id	int	Current Team ID
training_app_directory	str	Path to directory in Team File to training app default save directory
task_type	str	Model task type
widget_id	str	ID of the widget

team_id

Team ID where training app default save directory is located.

type: int

training_app_directory

Path to directory in Team File to training app default save directory. For example, if you have trained model in Train YOLOv8 application, you should provide /yolov8_train/ path.

type: str

task_type

Type of problem that model solves. Can be selected in the Training app GUI. For example, models that were trained with Train YOLOv8 application can have one of the following task types: object detection, instance segmentation, pose estimation.

type: str

widget_id

ID of the widget.

type: str

default value: None

Methods and attributes

Attributes and Methods	Returns
columns	List[str]
rows	Dict[str, List[ModelRow]]
get_selected_row()	ModelRow or None
get_selected_row_index()	int or None
set_active_row(index)	None
get_selected_task_type()	str
set_active_task_type(task_type)	None
get_available_task_types()	List[str]
get_selected_model_params()	Dict or None
use_custom_checkpoint_path()	bool
get_custom_checkpoint_name()	str
get_custom_checkpoint_path()	str
set_custom_checkpoint_path(path)	None
set_custom_checkpoint_preview(file_info)	None
get_custom_checkpoint_task_type()	str
set_custom_checkpoint_task_type(task_type)	None
enable()	None
disable()	None
enable_table()	None
disable_table()	None
@value_changed	Decorator function is handled when widget value is changed
@task_type_changed	Decorator function is handled when widget task type is changed

ModelRow

ModelRow object represents an automatically generated row in the CustomModelsSelector widget. It has information about the training session including:

• ID of the training session

• Names and paths to model artifacts

• Training data project

Function signature

ModelRow(
    api=api,
    team_id=team_id,
    task_id=task_id,
    task_path=task_path,
    training_project_name=training_project_name,
    artifacts_infos=artifacts_infos,
    session_link=session_link,
)

Methods and attributes

Attributes and Methods	Description
task_id	Return task id of the training session.
task_date	Return date and time of the training session.
task_link	Return link to the training session.
training_project_info	Return ProjectInfo object with project info.
artifacts_paths	Return list of all artifacts paths.
artifacts_names	Return list of artifacts names.
artifacts_selector	Return artifact selector widget with type Select.
get_selected_artifact_path()	Return path of the selected artifact.
get_selected_artifact_name()	Return name of the selected artifact.
to_html()	Converts object to html string to use it in the widget template.

Mini App Example

You can find this example in our Github repository:

supervisely-ecosystem/ui-widgets-demos/selection/019_trained_models_selector/src/main.py

Import libraries

import os
import supervisely as sly
from dotenv import load_dotenv
from supervisely.app.widgets import Card, CustomModelsSelector, Container, Text, Button

Init API client

First, we load environment variables with credentials and init API for communicating with Supervisely Instance:

load_dotenv("local.env")
load_dotenv(os.path.expanduser("~/supervisely.env"))

api = sly.Api()

Initialize CustomModelsSelector widget

team_id = sly.env.team_id()

checkpoint_infos = sly.nn.checkpoints.yolov8.get_list(api, team_id)
trained_models_table = CustomModelsSelector(team_id, checkpoint_infos)

Create additional widgets to preview selected model

model_name_preview = Text("", "text")
model_path_preview = Text("", "text")
preview_container = Container([model_name_preview, model_path_preview])
preview_container.hide()
preview_button = Button("Show preview")

Create app layout

Prepare a layout for app using Card widget with the content parameter and place widgets that we've just created in the Container widget.

container = Container([trained_models_table, preview_button, preview_container])
card = Card(title="TrainedModelsTable", content=container)
layout = card

Create app using layout

Create an app object with layout parameter.

app = sly.Application(layout=layout)

Add functions to control widgets from code

@trained_models_table.value_changed
def get_selected_row(row: CustomModelsSelector.ModelRow):
    preview_container.hide()


@preview_button.click
def preview_button_click_handler():
    preview_container.hide()
    row = trained_models_table.get_selected_row()
    model_name = row.get_selected_checkpoint_name()
    model_path = row.get_selected_checkpoint_path()

    model_name_preview.set(f"Model name: {model_name}", "text")
    model_path_preview.set(f"Model path: {model_path}", "text")
    preview_container.show()