multi-image_multi-semantic_lidar_late-fusion.yaml

_target_: opr.models.place_recognition.base.LateFusionModel

image_module:
  _target_: opr.models.place_recognition.base.ImageModel
  backbone:
    _target_: opr.modules.feature_extractors.ResNet18FPNFeatureExtractor
    in_channels: 3
    lateral_dim: 256
    fh_num_bottom_up: 4
    fh_num_top_down: 0
    pretrained: True
  head:
    _target_: opr.modules.GeM
  fusion:
    _target_: opr.modules.Add

semantic_module:
  _target_: opr.models.place_recognition.base.SemanticModel
  backbone:
    _target_: opr.modules.feature_extractors.ResNet18FPNFeatureExtractor
    in_channels: 1
    lateral_dim: 256
    fh_num_bottom_up: 4
    fh_num_top_down: 0
    pretrained: False
  head:
    _target_: opr.modules.GeM
  fusion:
    _target_: opr.modules.Add

cloud_module:
  _target_: opr.models.place_recognition.MinkLoc3Dv2
  in_channels: 1
  out_channels: 256
  num_top_down: 2
  conv0_kernel_size: 5
  block: ECABasicBlock
  layers: [1, 1, 1, 1]
  planes: [64, 128, 64, 32]
  pooling: gem

fusion_module:
  _target_: opr.modules.Concat