This is the first working TensorImage release, with several key features which will be further developed, with new ones added on later releases:
- __No code will have to be written__ by you when using TensorImage — making the entire process of training and classification more efficient — there are no errors that may arise while coding, allowing more time to be employed in feature engineering.
- __Automatically write file paths and labels__ for training dataset images.
- __TensorBoard visualization of training progress and model__ — making it easy to find out if hyperparameters still need to be tuned further and find out the best-performing model that you will use for classification.
- __Easily train image classification models__ — at the current stage there is just one CNN architecture available to use in TensorImage, however, this number will largely increase in future releases.
- __Readily classify thousands of images__ with image classification models trained with TensorImage — keeping what you need __compact and organized__.
- __Dataset preprocessing__ — at the current release you can only resize the datasets — this will be widely expanded in future releases.
- __Reusability__ — you can change your workspace directory at any time — TensorImage will not be affected by it — this is possible because all metadata and TensorImage outputs are stored in your workspace directory, which you can access at any time. You can have multiple workspaces for different image classification purposes,and change the workspace directory for TensorImage with no difficulty.