kitti.yaml

batch_size: 8
iters: 70000

train_dataset:
  type: KITTI
  dataset_root: datasets/kitti/training
  transforms:
    - type: Normalize
      mean: [0.485, 0.456, 0.406]
      std: [0.229, 0.224, 0.225]
  flip_prob: 0.5
  aug_prob: 0.3
  mode: train

val_dataset:
  type: KITTI
  dataset_root: datasets/kitti/training
  transforms:
    - type: Normalize
      mean: [0.485, 0.456, 0.406]
      std: [0.229, 0.224, 0.225]
  mode: val

optimizer:
  type: Adam

lr_scheduler:
  type: MultiStepDecay
  milestones: [36000, 55000]
  learning_rate: 1.25e-4