search

azhuantou / HSSDA

[CVPR 2023] Hierarchical Supervision and Shuffle Data Augmentation for 3D Semi-Supervised Object Detection
Apache License 2.0
68 stars 4 forks source link

PV-RCNN training results #23

Open JunjieWang0528 opened 11 months ago

JunjieWang0528 commented 11 months ago

I want to replicate your work, but the experimental results on the PV-RCNN model differ significantly, especially on pedestrians and bicycles. Can you provide me with your pre trained weight model or guide me on where I went wrong image

azhuantou commented 11 months ago

Hi @JunjieWang0528, thanks for your attention. Do you use the lastest code and could you show me the training log of your experiment for analysis?

JunjieWang0528 commented 11 months ago

I am using the latest code. Below is the training log of HSSDA under 1% data. There is a significant discrepancy between the results of pedestrian and bicycle inspections and your report image

log_train_20231218-093958.txt

azhuantou commented 11 months ago

Hi @JunjieWang0528 Following the 3DIoUMatch, the results reported in the paper are the averages of three different splits. If you only conduct experiments on one split, generating different results is normal.