Questions about the training -- TartanDrive missing?

[xzxzmmnn]

Thank you for providing the code. I have successfully trained the model from scratch using the public dataset you shared. The training process seemed to be successful; I've monitored the loss decrease and the visualization results in wandb appear to be satisfactory.

However, when I run the explore.py code using the model checkpoint file from my training, the model outputs are incorrect. To verify the functionality of explore.py, I also executed the code using the model checkpoint file you provided, which worked perfectly fine.

So, I am wondering if I need to modify the explore.py code to better utilize the model that I trained from scratch?

[liumingzs]

I want to know how many datasets you download,Can I only download the recon to train the model?

[ajaysridhar0]

@xzxzmmnn Can you describe how the model outputs are incorrect? @liumingzs You should be able to download all the datasets here. Let me know if you run into any specific issues.

[spokV]

It seems that TartanDrive is down. Nothing at http://airlab-share.andrew.cmu.edu/dataset-icra22/..

[robodhruv]

@spokV I recommend creating an issue at the relevant repo: https://github.com/castacks/tartan_drive or to the authors for assistance.

@liumingzs Yes, you should be able to train on a single (RECON) dataset too. But note that it may not generalize as well as a model trained on multiple datasets.