Are the models uploaded to HF trained or pre-trained?
chkp28 seems to have additional transformer layer between lstm and the output head which is not mention in the paper.
https://huggingface.co/datasets/raphael-blanchard/TartanIMU/tree/main/checkpoints/foundation_model
Are the models uploaded to HF trained or pre-trained?
chkp28 seems to have additional transformer layer between lstm and the output head which is not mention in the paper.
https://huggingface.co/datasets/raphael-blanchard/TartanIMU/tree/main/checkpoints/foundation_model