search

traveller59 / second.pytorch

SECOND for KITTI/NuScenes object detection
MIT License
1.72k stars 723 forks source link

Code on data augmentation #315

Open themousepotato opened 4 years ago

themousepotato commented 4 years ago

I would really appreciate if someone can point me to the data augmentation part in the codebase. Thanks!

xmyqsh commented 4 years ago

data_prep

darshb34 commented 4 years ago
  1. second.core.sample_ops.py contains the function sample_all().
  2. you can turn off the data augementation in second.builder.dataset_builder.py just comment below line in dataset_builder.py

if len(db_sampler_cfg.sample_groups) > 0 or db_sampler_cfg.database_info_path != "": # enable sample db_sampler = dbsampler_builder.build(db_sampler_cfg)

themousepotato commented 4 years ago

@xmyqsh @darshb34 Thank you very much for the information.

xmyqsh commented 4 years ago

db_sampler is the local augmentation that does not consume too much CPU time. There is also a global augmentation, its cpu time is directly proportional to the number of points.

But both of them are not the cpu time bottleneck, do you think so? @darshb34

abhigoku10 commented 4 years ago

@themousepotato @xmyqsh @darshb34 hi i have few queries on the data augmentation

  1. Can some one explain me what is GT_AUG , brief lines is good since i couldnot understand much from paper
  2. can we visualize the GT_AUG performed with the current code if so what is the command line

THanks in advance

Petros626 commented 1 month ago

@abhigoku10 Ground Truth Augmentation I guess.

Petros626 commented 1 month ago

@themousepotato did you succesfully tried data augmentation on the KITTI dataset?