mickeyzeng / data-visualization Goto Github PK

Status: Not Urgent
Description: Create a function to let user can upload model file and visualised the network model.

• Create an interface for user to upload
• Process the model file
• Display the model
• Test the function of visualisation

Some files have too many useless comments.

Extract the functions when needed, avoid to keep repeating.

Remove the unused methods, keep the code tidy and clean.

Remove unused console out to improve the speed.

Inspired by hamburger menu on code pen.

The reason for adding this menu is because there might be some features to add in the future but right now is nearly out of room - Maybe for switching between 2d and 3d display.

Status: Not Urgent
Description: The 3D display mode is a simple demo, which is needed to be upgraded.

• Display specific information (i.e., Kernal size, Output size, feature map)
• Complete the structure (i.e., residual Structure)

The function of CAM is implement the top 1 prediction. In the next, the user can click different prediction and get different CAM and the default setting is still top 1. Moreover, the layout will have specific sign to let the user know which prediction is selected now.

The current buttons are taken from Google Materialize Framework.

Customize the button looking is an add-ons.

Also some buttons have different functionalities but they look exactly the same - might be confusing to the user.

When added matplotlib.use() method, it only displays part of the feature map, if we do not add this function, the thread will be killed and Python will be terminated.

This issue is directly related to #1.

This function only works for CSV.

The goal is to load previous drew scribble back to the front-end.

• Upload network arch

• Upload network weights

• Upload class label

• Process in the backend

• Notice User if it is working

Negative pens have already finished. Now, the negative pens can let the neural network have different attention in an image.
Positive pens can make neural network have higher attention, which can make the neural network pay more attention to scribble area and have higher accuracy.

• Todo see here

Add a new function to submit the csv file which must be a legal format, otherwise, the system cannot process the file in right way. After submit the file, the system can have image preview, and user can click the submit to process the current image and check if there is any differences between the result of prediction and index in CSV file or not. If there are no difference, the prediction table would have special sign for the correct prediction.

The file should contain:

1. the URL(local address) of the image.
2. the index(the label) of image.

• Small bug needs to be fixed, see down below or click for more detail

Status: Urgent
Description: The current solution for generating feature map is using matplotlib to generate an image and then send to the front-end. It is massively slow and each time can only send one feature map.

• Change to multi dimension tensors
• Check feature map display on the back-end

Conclusion:

• Slider bar is not feasible, the speed is remarkably slow.
• The new way to generate feature map improved speed

Give user option to upload one image each time or a bunch of images within a folder.

If the user uses folder option, then, we provide two buttons for choosing next or previous image.

This thread will be using for all kinds of bug reports.

• CSV Blob does not have a file name - read more

• Current leader board gt result row background colour did not change when user clicked on it - read more

• The front-end can create two new global variables to make sure which type of JSON and network has been used by user.

• There is an issue about implementing customize network about GPU, which is needed to be discussed later.

User can select different colour map to show the different looking of feature map.

Add hint for the buttons when user uses mouse hover over the button.

As many buttons would confuse the users.

Also thinking about proper structure the first control panel area - the current one looks good but not ideal also see #16.

• Front-end Layout has subtle issues such as aspect ratio, part of the layout is hard coded which needs to be fixed.

• Design decision. There are some design decisions need to be discussed before implement. Such as where to select user's custom input model, where to download CSV, etc.

• Colour. The current interface is still not looking attractive. The colour matching can be improved.