search

JeffWang987 / OpenOccupancy

[ICCV 2023] OpenOccupancy: A Large Scale Benchmark for Surrounding Semantic Occupancy Perception
Apache License 2.0
574 stars 50 forks source link

The IoU of pretrained models is poor #32

Closed Junyu-Z closed 1 year ago

Junyu-Z commented 1 year ago

Thanks for sharing your code and pretrained models with us. I find the IoU of mutil-modal-baseline.pth/mutil-modal-CONet.pth/lidar-baseline.pth is significantly lower than the numbers reported in your paper. Is this phenomenon caused by the update of occupancy annotation? Looking forward your reply!

JeffWang987 commented 1 year ago

Apologies for any confusion. The low IoU values may be attributed to the shorter training iterations (15epc instead of 24epc) and the annotation update.

To clarify, the IoU values for mutil-modal-baseline.pth/mutil-modal-CONet.pth/lidar-baseline.pth should be 0.235/0.265/0.223 respectively. For further comparison, you can refer to the metric numbers in our training logs.

We will soon update these numbers in our paper.

Junyu-Z commented 1 year ago

Apologies for any confusion. The low IoU values may be attributed to the shorter training iterations (15epc instead of 24epc) and the annotation update.

To clarify, the IoU values for mutil-modal-baseline.pth/mutil-modal-CONet.pth/lidar-baseline.pth should be 0.235/0.265/0.223 respectively. For further comparison, you can refer to the metric numbers in our training logs.

We will soon update these numbers in our paper.

Congratulation on the acceptance by ICCV 2023! Recently, I'm following your work. And I would like to know when the final version paper is available.

Hub-Tian commented 8 months ago

Very wonderful work! Are there any updates on the release of your new paper? @JeffWang987