uber-research / LaneGCN

[ECCV2020 Oral] Learning Lane Graph Representations for Motion Forecasting
https://arxiv.org/abs/2007.13732
Other
507 stars 131 forks source link

Prediction error #32

Open WhoAmI-kai opened 2 years ago

WhoAmI-kai commented 2 years ago

I randomly selected 200 data from Argoverse Motion Forecasting's val for prediction testing, and the predictions were all nan.

The test command is: python test.py -m lanegcn --weight=36.000.ckpt --split=val Pre-trained models from: http://yun.sfo2.digitaloceanspaces.com/public/lanegcn/36.000.ckpt

image

The data of the input model is normal: image

But the predictions were all nan: image

Hope to get your help. @chenyuntc