Train your own AI Models
With Xailient Console, you can train custom AI models, which you can then deploy to your targeted devices.
To train any AI model, you will need training data.
Xailient Console comes with pre-loaded public datasets.
You can either use the pre-loaded datasets, upload your own dataset, or combine different datasets inorder to train your AI.
From the navigation menu on the left, go to MANAGE DATASETS.
Under Pre-loaded Public Datasets you will see a list of all the public datasets that we have pre-loaded and made available to you.
Upload your own Dataset
To prepare your images for uploading to the Console, place all the images into a folder. All images must be either in .jpg or .png format.
Now compress the folder into a .zip or .gz file.
Prepare labels for your dataset in CSV format. Please ensure your CSV has only these SIX columns. Spelling matters and it's case sensitive. The order doesn't matter.
E.g. ['image_name', 'class', 'xmin', 'ymin', 'xmax', 'ymax']
Here is what the CSV file should look like:
Go to the Console. From the navigation menu on the left, go to ADD NEW DATASETS.
Enter the name of the dataset, select the zip file yor prepared eariler and click UPLOAD.
When the upload begins, do not close the browser to ensure that the dataset is uploaded successfully.
Now upload your label file you prepared earlier.
When the lable file upload is completed, you will then get redirected to MANAGE DATASETS page where your dataset is currently getting sanity checked before it becomes available. This can take a few minutes. Refresh the page if it takes longer than 20 minutes!
If the sanity check passes without errors, it will get automatically listed to the above list and will now be ready to train on.
If the sanity check fails then the ERROR DETAILS could explain the reason. For example, here we have uploaded a CSV labels file where not a single image in the .zip file has a corresponding annotation. So make sure at least 1 row in the CSV file corresponds to an image filename in the .zip file. You can correct these errors and re-upload the CSV by clicking UPDATE LABEL.
Sometimes the sanity check will pass with partial success because we detected that some images don’t contain labels and vice versa, which is acceptable since this means the rest have corresponding labels. In this case you acknowledge the error and can manually approve it by clicking the APPROVE button.
This approved dataset will then get listed above and is now ready to train on.
Train your AI
To train an AI model, navigate to TRAIN NEW AI MODEL from the menu on the left.
Select dataset(s) by ticking the box(es). You can select a single or combine multiple (both public and private) datasets to train a model.
You can further select only specific subclasses within the (set of) datasets. A single model will train on the combined datasets and predict only those classes you have chosen here.
Enter model name for your AI model.
Select a value for Speed vs Accuracy between 1 to 5.
1 is the fastest. It is best for larger objects with simple backgrounds. You should select option 1 if your objects of interest are medium to big (close range).
2 is slower than 1 and should be used if your objects of interest are small to medium (medium range).
3 is slower than 2 and should be used if your objects of interest are small (long range).
4 is slower than 3 and should be used if your objects of interest are smaller and background is complex (longer range).
5 is the slowest. It is best for smaller objects with complex backgrounds. You should use option 5 if your objects of interest are very small and background is complex (longest range).
Begin training by clicking BEGIN TRAINING.
View Training Status
To view the training status of your AI model, navigate to MANAGE AI MODELS, and find your model under Training Models. Once the training is completed, you will get an email notification regarding it.
Deploy your AI
When the training is complete, you can build an SDK and install it on the targeted device. For more details on how to build an SDK, refer to Build Deployable SDK section of the documentation
For instructions on how to setup your device, install the SDK and run it, please refer to Deploy You AI section of the documentation.
If your local machines doesn't meet the requirements for native installation we also offer Docker Containers as a deployment option.