Draft: [Fix] Increase precision of exported float parameters
Context
As mentioned in this issue, the backend_cpu and export_cpp float results diverge at some point.
Changes
Comparison was made using the aidge_cmp export feature, which compares the export feature maps with the ones we get from the aidge_backend_cpu.
Aidge compare tests :
| model | fp32 | int8 |
|---|---|---|
| lenet | ||
| resnet18 | ||
| vgg16 |
Edited by Axel Farrugia