After logging in to your account, you will be able to start by loading the training data-set into DLHUB
In the LOAD TRAINING DATA page, use the browse button to select your training data-set file/folder
After loading the data, up to 100 sample items will be shown for a user to review the data.
Total Number of Samples and Number of Categories will be indicated in the file summary section.
Please review and reorganize your data/folder if displayed summary is incorrect
If loaded data is not image type, please un-tick "Please enable if it is Image Data" tick-box and press NEXT to move on to the next process. If loaded data is of image type (file or folder), please tick the "Please enable if it is Image Data" tick-box and continue the following steps.
Select Model Type and Input Shape
Select a pre-trained model if you wish to use Transfer Learning. (Refer to Transfer Learning section for more details)
By choosing a Transfer Learning model, the Input Shape array will be automatically populated.
For example, if you choose ResNet18_ImageNet_CNT, Input Shape will be auto populated as 224x224x3 as that's the image shape used during its training.
It's OK if the pre-trained model Input Shape is different to your own data-set, DLHUB has embedded auto-scaling functions to adapt your data to use the pre-trained models.
Choose User_Defined_Model if you wish to configure your own model.
user will then have to define the Input Shape based on the image: Width (pixel) x Height (pixel) x Depth (bits).
For User_Defined_Model, DLHUB will check will check if the input shape matches your loaded data. Green means good, and Red means there is a mismatch. Please review your data or Input Shape and make sure they match. You won't be able to proceed if the check is indicated as BAD.
Tick: if you wish you normalize your input image, this will often enhance deep learning outcome.
Untick: if you don't want to perform image normalization