search

woodfrog / maptracker

Code for paper "MapTracker: Tracking with Strided Memory Fusion for Consistent Vector HD Mapping", ECCV 2024 (Oral)
https://map-tracker.github.io/
Other
144 stars 10 forks source link

trans_loss error and pred.shape[0]==0 #13

Closed yj772881654 closed 2 months ago

yj772881654 commented 2 months ago
image image

During training, I encountered an error where the pred shape was 0, and the trans loss returned a None error. I have made modifications to get the correct results, but I would like to ask if you have this problem over there. What is the reason?

woodfrog commented 2 months ago

I don't remember getting exactly the same error in my own training. Did you get this error at the first few iterations of the training? If yes, I can guess the reason:

At the very beginning of the training, it's possible that all the outputs have low confidence and no prediction is propagated to the next frame. When this happens, the tracked query has length 0 and will trigger the error you encountered.