jychoi118 / ilvr_adm

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models (ICCV 2021 Oral)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About super_res_sample

zxt-triumph opened this issue · comments

What is the base_samples?

It is directory to the low-resolution input images.

When I specify the directory, I get the following error:

  • python scripts/super_res_sample.py --attention_resolutions 16 --class_cond False --diffusion_steps 1000 --dropout 0.0 --large_size 256 --small_size 64 --learn_sigma False --noise_schedule linear --num_channels 128 --num_head_channels 4 --num_res_blocks 2 --resblock_updown False --use_fp16 False --use_scale_shift_norm False --timestep_respacing 100 --model_path ./logs/res_256/ema_0.9999_002000.pt --base_samples ./datasets/test/
    Logging to ./logs/res_256/
    creating model...
    loading data...
    creating samples...
    Traceback (most recent call last):
    File "/home/ztf/project/zxt/Code/Diffusion Models/Noise2Noise/scripts/super_res_sample.py", line 119, in
    main()
    File "/home/ztf/project/zxt/Code/Diffusion Models/Noise2Noise/scripts/super_res_sample.py", line 47, in main
    model_kwargs = next(data)
    File "/home/ztf/project/zxt/Code/Diffusion Models/Noise2Noise/scripts/super_res_sample.py", line 78, in load_data_for_worker
    with bf.BlobFile(base_samples, "rb") as f:
    File "/home/ztf/anaconda3/envs/vgan/lib/python3.9/site-packages/blobfile/_ops.py", line 356, in BlobFile
    return default_context.BlobFile(
    File "/home/ztf/anaconda3/envs/vgan/lib/python3.9/site-packages/blobfile/_context.py", line 859, in BlobFile
    raise IsADirectoryError(f"Is a directory: '{path}'")
    IsADirectoryError: Is a directory: './datasets/test/'

Oh, in this script, base_sample is .npz file of low-resolution images.
You can modify the script by using data loader from super_res_train.py.
For example,
image
image