search

antoniorv6 / SMT

Official implementation of the Sheet Music Transformer
MIT License
5 stars 6 forks source link

error when predict with python transcribe_single_score.py #4

Closed lvpchen closed 3 months ago

lvpchen commented 5 months ago

python transcribe_single_score.py --config config/GrandStaff/CNNSMT.gin --sample_image Data/GrandStaff/grandstaff_dataset/beethoven/piano-sonatas/sonata01-1/original_m-0-5.jpg --model_weights weights/GrandStaff/CNNSMT.ckpt

Sometimes, there is an error when i run predict.

File "SMT/model/Decoder.py", line 28, in forward return x + self.pe[:, :, start:start+x.size(2)].to(x.device) RuntimeError: The size of tensor a (833) must match the size of tensor b (832) at non-singleton dimension 2 In call to configurable 'main' (<function main at 0x7f473ff3ef70>)

antoniorv6 commented 3 months ago

Seems that the bug is happening during prediction, probably you are putting in the decoder more tokens than it is able to produce. I will check it out once the rework of the repository has been finished.

antoniorv6 commented 3 months ago

Closing this issue, as the whole code of the SMT has changed to be more straightforward. Please, notify me if errors happen when using this new version.