SR yml is missing?

[liulanyang-alpha]

SR yml is missing?

[Amirhosein-gh98]

SR option file is added now.

[njzxj]

latent_attn_type1 : "CTS"
latent_attn_type2 : "Channel"
latent_attn_type3 : "CTS"
latent_ffw_type : "GFFW"

In turtle_arch.py CTS Not defined

[Amirhosein-gh98]

These were our old namings. I fixed the issue by renaming the modules in the option file.

[njzxj]

When I set the following parameters:

config = "Turtle/options/Turtle_SR_MVSR.yml"
model_path = "models/SuperResolution.pth"

I encounter the following errors:

Error(s) in loading state_dict for Turtle:
    Unexpected key(s) in state_dict: "decoder_level3.transformer_blocks.9.attn.spatial_aligner.k2.weight", "decoder_level3.transformer_blocks.9.attn.spatial_aligner.k2_dwconv.weight", "decoder_level3.transformer_blocks.9.attn.spatial_aligner.q2.weight", "decoder_level3.transformer_blocks.9.attn.spatial_aligner.q2_dwconv.weight", "decoder_level2.transformer_blocks.5.attn.spatial_aligner.k2.weight", "decoder_level2.transformer_blocks.5.attn.spatial_aligner.k2_dwconv.weight", "decoder_level2.transformer_blocks.5.attn.spatial_aligner.q2.weight", "decoder_level2.transformer_blocks.5.attn.spatial_aligner.q2_dwconv.weight", "decoder_level1.transformer_blocks.1.attn.spatial_aligner.k2.weight", "decoder_level1.transformer_blocks.1.attn.spatial_aligner.k2_dwconv.weight", "decoder_level1.transformer_blocks.1.attn.spatial_aligner.q2.weight", "decoder_level1.transformer_blocks.1.attn.spatial_aligner.q2_dwconv.weight". 
Size mismatch for decoder_level3.transformer_blocks.9.attn.spatial_aligner.temperature: copying a parameter with shape torch.Size([1, 1, 1]) from checkpoint, while the shape in the current model is torch.Size([4, 1, 1]).
Size mismatch for decoder_level2.transformer_blocks.5.attn.spatial_aligner.temperature: copying a parameter with shape torch.Size([1, 1, 1]) from checkpoint, while the shape in the current model is torch.Size([2, 1, 1]).

The error occurs at line 121 in /root/code/Turtle/basicsr/inference_no_ground_truth.py, where model.load_state_dict(torch.load(path)['params']) is called.

Is there an issue with the configuration file or something else?

[zelenooki87]

Having exactly same issue!

[zelenooki87]

solved:
Changed model_type="t0",
to
model_type="t1",

[Amirhosein-gh98]

The model type should be changed to SR for Super Resolution. Could you elaborate more on what you did and for which task?
An upsampling was also missing for the SR case in the inference_no_ground_truth.py file, which is fixed now.