AssemblyAI-Community/MinImagen

Noise Image generation as a result of inference

Opened this issue · 2 comments

evg-cv commented

I followed the instructions for training the model and generation of images using inference.py.
That is,

  1. git clone https://github.com/AssemblyAI-Examples/MinImagen/
  2. pip install -r requirements.txt
  3. python train.py --PARAMETERS ./parameters --BATCH_SIZE 2 --TIMESTEPS 25 --TESTING
  4. python inference.py -CAPTIONS captions.txt --TRAINING_DIRECTORY training_

So finally, I could get the generated images for 2 captions: 'a happy dog' and 'a big red house', but all of them are nosied as attached.
image_0
image_1

Though I followed the same instructions as README without any modification, very noised images were generated. What's the reason?

Guhanxue commented

Same here; I also only get noise, and I also tried to check their repo and blog but didn't see any generated image results.

B00912163 commented

I got the same issue. Does anyone got the solution for this?