2. Design and Planing

Initial Design and Planing for Deep Images Hub(DIH)

As an Images Supplier, I am able to upload an batch of images with just a single label.

Simulate batch images submission by copying image sources from S3 buckets to DIH's S3 buckets
DIH shall have the mechanism to verify user submitted labels if exist in DIH's database.
- If label doesn't exist inform user.
DIH shall have the mechanism to chain the user supplied label into its own branch of category
- Create backend data relationship to organize and allocate data
If user opt-in to share their location of where they uploaded the images, DIH shall be able to record that.
- Simulate user geolocation info

As a Data Scientist, when I visit Deep Image Hub I should see how many categories(labels) of images I can download.
As a Data Scientist, when I visit Deep Image Hub I should see the latest batches of images that just uploaded.
As a Data Scientist, I should be able to select labels of images and download the images by categories.
- For example, if I choose the category Food, I should be able to download all the images about Food.
As a Data Scientist, I should be able to request new label if it is not already in DIH
As a Data Scientist, I should be able to request to train a baseline model with my choice of labels of images.
- And if there are not enough images (above 500) for certain labels, I can still enqueue my training request
As a Data Scientist, I should be able to get a the download link of my model once it is trained.
- The model training summary about the final accuracy scores and losses should be reported as well

DIH webpage shall display all the labels that has images and they should be able to group by their own categories
- The label name and the number of images under this label shall be displayed
DIH webpage shall be able to constantly display the latest image batches submissions with label name, number of images and where they came from
DIH shall allow user to download batch of images by the parent category of the images
- Building label relationships and use hierarchical and recursive queries in SQL to achieve this request
DIH shall able to allow user to add new labels once they also provide the immediate parent label
- For example LaCorix's immediate parent label will be soft_drink
DIH shall have a user requests watch list to keep track of the user requests
- A schedule workflow will be needed to constantly check if the requirements are full filled
DIH should keep track of the modeling training results and display on the model list web page
- A email with downloadable links and brief summary of the model training report should send to user once model training is done.

Provide feedback