nicecv / DiffHarmony

The official pytorch implementation of DiffHarmony and DiffHarmony++.
7 stars 3 forks source link

Some help to launch a simple evaluation on an image #2

Open the-ponpon opened 3 weeks ago

the-ponpon commented 3 weeks ago

Hi there,

First of all thank you for sharing your work.

I have downloaded the weights from the Google drive, thank you!. Now I am having some difficulties to understand how I pass the weights to launch a simple evaluation on an image. A command line showing how to call the inference and how to pass the weights information would be very helpful.

Thank you in advance,

Pierre

rdancer commented 2 weeks ago

@the-ponpon

  1. Download models
  2. Change inference.sh
  3. Create test subdirectories and populate them with specially named files
  4. Create test.jsonl
  5. Run inference.sh
  6. Collect output files from the output directory

Basically you need to do everything I have done in my fork, in this commit -- read the whole commit, especially the data/README.MD file

There is also src/utils.py::make_comp() but that needs the pipeline and is currently not called from anywhere and looks unfinished. This will be the same pipeline that is currently only constructed when calling scripts/inference.sh which in turn calls scripts/inference/main.py which constructs the pipeline, so if you want to call it from Python and without the extra overhead, you will need to copy some of the code from main.py, populated with the configuration from inference.sh.