Input, Output Data / Value Testing Feature

[struss]

Purpose

Let's support this feature.

Input & Output Data

• [ ] Create data for each model ( Currently on npy file) (end of APR)
  • [ ] Support multiple type(f32, Q8, I16...) (end of MAY)
  • [ ] Support multiple format(npy, hd5, binary...) (end of MAY)
  • Adding Convertor for this. (like npy -> hd5 or npy -> binary) (end of MAY)
  • [ ] use on every compile step(end of MAY)
• [ ] Fix uniform format for input & output (APR)
• [ ] Image data handling (?)
  • [ ] Input data pre-processing : Image to npy file(etc) or more......
  • [ ] output data processing : output file to Image (maybe?)

Value Testing

• [ ] Value test for multiple model on multiple steps
  • [ ] Prepare tool or method for multiple model (Where to Run inference?) #364
  • [ ] TFlite : Use TFLite module in inference by script
  • [ ] pytorch : Use pytorch module in inference by script
  • [ ] circle : Use <WHAT> in inference by script?
  • [ ] NPU : Use <WHAT> in inference by script?
  • [ ] ...
• [ ] Visualize difference between compile steps (?)
• [ ] Image output compare tool (end of June)

#

Please feel free to add any contents about this topic.

[struss]

Before make a testing, we need to fix format for input & output.

[YongseopKim]

How about adding the deadline for each task?

[struss]

How about adding the deadline for each task?

Oh I will append that! thanks!

[YongseopKim]

I'm interested in Input & Output Data without Image data handling.

[hyunsik-yoon]

@struss,

Visualize difference between compile steps

I am not sure what this task is. Could you elaborate more about this?

[struss]

Talk about user who use ONE as a main tool needed feature about this : only value test feature first like PEIR or mPEIR MRIA result will work nice on this

after those value check passed, bounding box or some other visualization would be nice.

# So, first target of this visualization is PEIR value and MRIA result graph