I am trying to implement the sparseocc to occupancy flow challenge (cvpr 2024), but the training result is not good, below is the result after 24 epochs on the validation dataset.
The details of the implementation simply replace the GT with the latest challenge GT, change the class number (18->17), and do not consider the flow training.
Have you tried to train on the latest challenge data? Thanks for any comments and suggestions.
GANWANSHUI
commented
7 months ago
hi, thanks a lot for the nice work.
I am trying to implement the sparseocc to occupancy flow challenge (cvpr 2024), but the training result is not good, below is the result after 24 epochs on the validation dataset.
The details of the implementation simply replace the GT with the latest challenge GT, change the class number (18->17), and do not consider the flow training.
Have you tried to train on the latest challenge data? Thanks for any comments and suggestions.
| Class Names | RayIoU@1 | RayIoU@2 | RayIoU@4 | +----------------------+----------+----------+----------+ | car | 0.103 | 0.152 | 0.182 | | truck | 0.044 | 0.077 | 0.107 | | trailer | 0.024 | 0.038 | 0.059 | | bus | 0.062 | 0.113 | 0.199 | | construction_vehicle | 0.054 | 0.094 | 0.143 | | bicycle | 0.000 | 0.000 | 0.000 | | motorcycle | 0.000 | 0.000 | 0.000 | | pedestrian | 0.000 | 0.001 | 0.001 | | traffic_cone | 0.000 | 0.000 | 0.000 | | barrier | 0.021 | 0.028 | 0.035 | | driveable_surface | 0.112 | 0.164 | 0.225 | | other_flat | 0.005 | 0.008 | 0.010 | | sidewalk | 0.058 | 0.091 | 0.126 | | terrain | 0.064 | 0.106 | 0.157 | | manmade | 0.034 | 0.054 | 0.085 | | vegetation | 0.020 | 0.036 | 0.063 | +----------------------+----------+----------+----------+ | MEAN | 0.038 | 0.060 | 0.087 | +----------------------+----------+----------+----------+