search

Vegeta2020 / SE-SSD

SE-SSD: Self-Ensembling Single-Stage Object Detector From Point Cloud, CVPR 2021.
Apache License 2.0
811 stars 128 forks source link

How to extend the effective detection range #65

Closed Jason-wwww closed 2 years ago

Jason-wwww commented 2 years ago

In order to extend the effective detection range, I modify the value of "target_assigner", "test_config" and "voxel_generator" in the config file.

anchor_ranges=[0, -40.0, -1.0, 70.4, 40.0, 1.0] ->anchor_ranges=[0, -40.0, -1.0, 102.4, 40.0, 1.0] in "target_assigner"

range=[0, -40.0, -3.0, 70.4, 40.0, 1.0] -> anchor_ranges=[0, -40.0, -3.0, 102.4, 40.0, 1.0] in "voxel_generator"

post_center_limit_range=[0, -40.0, -5.0, 70.4, 40.0, 5.0] -> anchor_ranges=[0, -40.0, -5.0, 102.4, 40.0, 5.0] in "test_config"

However, the effective detection range is also about 70m. What can I do to extend the effective detection range?

Vegeta2020 commented 2 years ago

@Jason-wwww Do you test it on KITTI or a new dataset? If on KITTI, there may exist very few points out of 70m range. If on a new dataset, I wonder if there exist enough points for the long-distance objects. For your revisions, I think they are right, while I'm not sure if they are enough as I don't remember too much, but I suggest you to check more on the results before NMS -> then the confidence maps of the detection head, by this way, you should find out how to address it.