julrog / nn_vis Goto Github PK

Add Documentation for installation and requirements.

For making the tools work for more types of neural networks the shaders/code that are until now fixed on 10 output classes of a neural network instead dynamic.

• dynamic render shader
• dynamic render shader loading
• dynamic compute shader
• dynamic compute shader loading
• dynamic buffer
• testing

Introduce a FPS limit, to reduce the load on the GPU, while simply rendering.

Rework processing using specified configuration files for processing and rendering. Also add a new config for the automation.

Hi there, this is a wonderful work, I can see model directly, I have not try but I wonder if this work support Pytorch model? if not, will it be supported in the future?

As the title suggests, after running the start_tool.py--demo file, the model structure is not visualised

Code:
from data.mnist_data_handler import split_mnist_data

Error message:
C:\Users\Yinqi Huang\nn_vis>python examples/process_mnist_model.py Traceback (most recent call last): File "C:\Users\Yinqi Huang\nn_vis\examples\process_mnist_model.py", line 4, in <module> from data.mnist_data_handler import split_mnist_data ModuleNotFoundError: No module named 'data'

Refactor large files and functions, which are too complex.

Refactored Folders:

• data
• evaluation
• gui
• models
• neural_network_processing
• opengl_helper
• processing
• rendering
• examples
• utility
• root

The processing can take some time. If it needs to stop or an error occurs and crashes, it would be good to have a save file on which you can continue processing without restarting everything.

As title, anyway to import tensorflow.keras pretrained model file ? can it run to show visualization ?

在load processed network时，似乎只能加载.pro.npz同时存在的文件，是否能让.pb文件加载进来呢？如果要加载.pb文件是否要编写对应.pb文件的脚本来解析其中的内容？

Closing the GUI window doesn't close the second window used for rendering the neural network. This should be fixed and tested. Also the threads should be checked to close all together.

Reset values a user puts in the GUI, if they are not accepted.

Not all neural networks are supported. Add checks for valid layer types and other attributes of a neural network to check its validity.

Look up good documentation style and apply it to the whole code.

Encapsulate configurations as one or multiple files (depending on the different tools) and add save/load functionality.

Create a fancy site showcasing processed networks and evaluation data form the thesis.

Create a config file, which will be stored and overwritten for window positions of the conrtol panel and the render window.