📋 [TASK] Geti Inspect: Improve train model dialog #3175
Description
Describe the task
The current training dialog is a bit overwhelming in the amount of models the user can choose between. In addition we don't guide the users yet to choosing the correct model.
To improve this we should get some more info about the available models including
- id / name destinction (to avoid showing
snake_casenames in the UI) - filter out old models
- add 3 recommended models
- add relative metric scores for training, inference and accuracy
- add links to documentation
The relative metric scores are still up for discussion as we'd need to be able to cross reference to the benchmarks that @rajeshgangireddy ran.
Update /trainable-models with extra semantic information
{
id: string; // id that is returned at this moment
name: string; // human readable name, I believe anomalib has some functions to output them in "Title case"?
recommended: boolean; // the three models Ashwin and Rajesh are recommending. This is optional, we could also hardcode it on the UI side (we have to hardcode it somewhere anyway)
docs: string; // (maybe) an option to direct the user to the docs
metrics: { // (maybe)
training: ScoreMetric;
inference: ScoreMetric;
accuracy: ScoreMetric;
},
}
Show only 3 recommended models by default
Hide the other models under a "Show all" button.
Show the training device at the bottom, after selecting the model
By default select a GPU device.
Hide winclip and video models
These are not yet supported by our application logic.
Acceptance Criteria
See task description
Priority
Low
Related Epic
No response
Estimated Time
No response
Current Status
In Progress
Additional Information
No response