Getting Started Guide
- Model development using Notebook/Template
- Machine learning process guide
- work flow
- Screen Description
- Creating a user
- Create a Project
- Upload your data
- Create a work
- Create Tasks
- Assign annotators to a work
- Perform annotation work
- Review the work
- Output the results
- Custom Image
ABEJA Platform CLI
- CONFIG COMMAND
- DATALAKE COMMAND
- DATASET COMMAND
- TRAINING COMMAND
- startapp command
- ABEJA Platform CLI
Template Training (Object Detection)
This page explains how to use ABEJA Template to create Object Detection models without coding.
This template uses the Object Detection data set. Prepare the Object Detection data set referring to the following.
■Create a sample dataset
You can also create a PASCAL VOC sample dataset using this script. (This is an image dataset with annotation data.)
■Upload a file to the datalake and create a dataset using the annotation tool
■If you already have annotated data
Create a machine learning model using ABEJA Template based on the dataset.
First, create a job definition from the “Learning” “Job Definition” page of the console.and Select “No sample”.
Then create a version.After selecting “Create Version”, Click “Template” on the tab, and select Object Detection CPU or GPU. In this example, select GPU. You can also adjust environment variables (Hyperparameters) on this page. Please refer here for the explanation of various hyper parameter information.
“Job version” has been created in the job definition.
Next, click “Job” to create a learning job.
Set the following items on this screen.
|Job definition version||Select version to use for learning|
|Instance type||Select each instance type (Select
|Dataset||Specify the name of the dataset to be trained by the model. For the alias, enter train or Val|
|Environment variable||Displays various parameters defined by the job version. Each item can be adjusted by selecting the edit button on the right|
Finally, select Create Learning Job to run the learning job. A learning job was executed. Parameters and data sets set for each job can also be checked on the screen.
When you click the TensorBoard button after learning starts, the TensorBoard screen opens and you can visualize the training loss and validation loss data.
The progress of learning is displayed on the management screen, so wait for the learning to finish.
Here, we explained how to create a model for Object Detection without coding using ABEJA Template.