aofrancani/TSformer-VO

Using pre trained models.

Closed this issue · 3 comments

Hi.
I am trying to run the pre-trained models on a KITTI sequence (03). Predict_poses works fine but outputs an empty .npy file. I'm confused as to why this happens. Any help would be appreciated!

Do I have to train the model first and then run predict_poses.py? I just need some clarification.

Hey, what do you mean by "predict_poses" works fine? Maybe you are not setting the right path do your pretrained model, I'm not sure... if it works fine you should have non-empty .npy files.

Yes, you can either train a new model from scratch or you can download one of my pretrained models here: https://github.com/aofrancani/TSformer-VO?tab=readme-ov-file#2-pre-trained-models
Then try checking if you are loading it correctly (https://github.com/aofrancani/TSformer-VO?tab=readme-ov-file#42-inference)

Another possibility is that you are not setting the correct path to your dataset. I forgot to use a parameter in this code to set the "data_path", so you are loading the data considering the default path.
Try changing it here (https://github.com/aofrancani/TSformer-VO/blob/main/datasets/kitti.py#L23) or insert an input parameter "data_path" in predict_poses.py (https://github.com/aofrancani/TSformer-VO/blob/main/predict_poses.py#L61)

I hope this helps, and I'm sorry for that!

I figured it out. I don't have CUDA on my system so it was actually skipping the pose predicting part. I changed the code accordingly and now it works perfectly. Thanks for your help and your work on this repo!