search

nodefluxio / vortex

A Deep Learning Model Development Framework for Computer Vision
27 stars 6 forks source link

[FEATURE] Additional attributes for vortex model #5

Closed alphinside closed 4 years ago

alphinside commented 4 years ago

Is your feature request related to a problem? Please describe. When using prediction pipeline, the behavior of vortex model is different compared to vortex runtime model.

  1. It doesnt have any 'input_specs' attribute
  2. It required dataset to be downloaded to get the class_names attributes

Additionally, we also lack of information to resume interrupted training

So, we need this to be implemented

Describe the solution you'd like Currently, .pth is directly stored the model's params. We can extend this to store a dictionary of model's attributes and training states

  1. Adding input specs
  2. Adding class_names
  3. Adding current epoch and optimizer state
  4. Update scheduler state with info from ckpt on resume