search

brade31919 / radar_depth

Source code of the IROS 2020 paper "Depth Estimation from Monocular Images and Sparse Radar Data"
MIT License
87 stars 13 forks source link

About train / val scene numbers #8

Closed zzangjinsun closed 3 years ago

zzangjinsun commented 3 years ago

Hi @brade31919 ,

I have a quick question about scene numbers for train / val splits of the released preprocessed dataset.

  1. Does the dataset use the exact scene numbers for train / val splits from the following codes?
  2. The random seed is always fixed to 100. Did you also use the seed 100 for the dataset?

https://github.com/brade31919/radar_depth/blob/5e6e75772ff379aac65379a50d4042a7c64c869d/dataset/nuscenes_dataset.py#L116-L128

Thank you very much!

brade31919 commented 3 years ago

Hi @zzangjinsun,

  1. Yes, the processed dataset was created by the code in that file. Since we haven't made sure that the code for creating the processed is ready, we didn't mention it in the README.
  2. Yes, I didn't change the random seed. You can check whether the dataset splitting results match the contents in the processed dataset from your side (by checking the images). I am not sure whether np.random.choice will have consistent results across different library versions...

Best

zzangjinsun commented 3 years ago

I manually checked and concluded that the code is matched with the preprocessed dataset. Thx!