search

MCG-NJU / SparseOcc

[ECCV 2024] Fully Sparse 3D Occupancy Prediction & RayIoU Evaluation Metric
https://arxiv.org/abs/2312.17118
Apache License 2.0
277 stars 22 forks source link

Change the evaluation metrics to miou #62

Open autumn-wong opened 1 week ago

autumn-wong commented 1 week ago

Currently there are very few occupancy networks based on sparse frameworks, thank you for your work, which has inspired me to be able to build on it to be able to continue with new ideas in my research. I tried to change the evaluation metrics to miou to be able to easily compare the performance with other models, but I noticed that running 'python old_metrics.py' needs to save the predictions in npz format. What needs to be changed in the code to be able to use miou when executing val.py? Can you help me in this matter?

YANG-CY-163 commented 6 days ago

You can rewrite the evaluate method of nuscenes_occ_dataset.py to use Metric_mIoU instead of main_rayiou; Or you can simply add the following code after getting the sem_pred with shape [200x200x16] in the evaluate method, saving predictions when executing val.py. image

autumn-wong commented 6 days ago

I will try it. Thanks for your quick reply!