cene555 / imagen-pytorch Goto Github PK

restarted non pro colab runtime clean but error persists and fails at[16]?
face_enhancer = GFPGANer(
model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
upscale=4,
arch='clean',
channel_multiplier=2,
bg_upsampler=upsampler)

RuntimeError Traceback (most recent call last)
in ()
4 arch='clean',
5 channel_multiplier=2,
----> 6 bg_upsampler=upsampler)

2 frames
/usr/local/lib/python3.7/dist-packages/torch/serialization.py in _legacy_load(f, map_location, pickle_module, **pickle_load_args)
938 typed_storage._storage._set_from_file(
939 f, offset, f_should_read_directly,
--> 940 torch._utils._element_size(typed_storage.dtype))
941 if offset is not None:
942 offset = f.tell()

RuntimeError: unexpected EOF, expected 7441746 more bytes. The file might be corrupted.

Hi,

How do I upsample the results that I'm seeing with the notebook?

Generation result looks not so good...
For example:

prompt = 'monkey eat banana'

Maybe someone get better generation result?

The paper stated they use different number of resnet blocks for super resolution, but the impl in this repo doesn't reflect that or I miss something?

#1

That model is a quick training in colab, don't expect good results

Maybe this could be made clearer in the README? @cene555

Otherwise, I think it just wastes people's time if all they're trying to get is something to use like dalle-mini (which is also limited, but arguably much better than the one linked in Colab), but at the expense of needing Colab Pro or similar.

Thank you for publishing this great code.
I marvel at the excellence of this code and would love to use it.

What is the license for this code?

May I use it commercially, like the MIT licence?

from celli:
i add this to the bottom of cene notebook for inference

https://gist.github.com/rom1504/b5db3c98c5485c0ec5a1d22d79ca083a

#benchmark
import random
import re
from shutil import copyfile

prompts2=[]
with open('/content/bench.txt') as my_file:
    prompts = my_file.readlines()

for q in prompts:
  q=q.strip()
  q=q.replace('\n', '')
  prompts2.append(q) 

prompts=prompts2
print(prompts[0])

for bench in prompts:
  prompt=bench
  text_encoding = tokenizer(
    prompt,
    max_length=128,
    padding="max_length",
    truncation=True,
    return_attention_mask=True,
    add_special_tokens=True,
    return_tensors="pt"
  )

  uncond_text_encoding = tokenizer(
      '',
      max_length=128,
      padding="max_length",
      truncation=True,
      return_attention_mask=True,
      add_special_tokens=True,
      return_tensors="pt"
  )
  model_kwargs = {}
  model_kwargs["tokens"] = th.cat((text_encoding['input_ids'],
                                  uncond_text_encoding['input_ids'])).to(device)
  model_kwargs["mask"] = th.cat((text_encoding['attention_mask'],
                                uncond_text_encoding['attention_mask'])).to(device)

  sample = diffusion.p_sample_loop(
      model_fn,
      (2, 3, 64, 64),
      clip_denoised=True,
      model_kwargs=model_kwargs,
      device='cuda',
      #progress=True,
  )[:1]
  print(prompt)
  show_images(sample)

NOT ENOUGHT RAM IN COLAB , can i run it localy on rtx 3090 on windows ? how much ram its need?