search

rd20karim / M2T-Interpretable

[BMVC 2024] Official Implementation of the paper guided attention for interpretable motion captioning
MIT License
6 stars 0 forks source link

Cuda error: CUDA error: device-side assert triggered #2

Open Stefano-retinize opened 2 weeks ago

Stefano-retinize commented 2 weeks ago

Hey @rd20karim thsnks for your help do far, I am having this error in the code line: Line 78

It seems it has to do with the number of classes the model is predicting or something.

Maybe related with that is that I noticed that the vocab size inside vocabulary.vocab_size_unk is different than the KIT model input_dim. So, I guess somehow the vocabulary class is building a longer vocabulary than what the trained model have.

Is there any chance you can share the class vocabulary object maybe as a .pkl (for the models model_ton5mfwh and model_u8qatt2y)? Maybe this is faster, since I guess is going to be more difficult to spot what's the error when is building this vocabulary class.

Thanks for your help.!

rd20karim commented 2 weeks ago

Hi @Stefano-retinize, Did you download KIT-ML or HumanML3D dataset? what is the size of your current vocabulary. Building the dataset should be done using only the scripts in the other project code

Stefano-retinize commented 2 weeks ago

Yeah I actually changed to the segmentation project and it all worked fine. I am now trying to figure out what each number means in the dataset. I think is all position based.

Thanks for all your help!