name: feature_extractor_v2
backbone:
  encoder: resnet50
  pretrained: true
  output_dim: ${...latent_dim}
  num_downsample: null
  remove_stride_from_first_conv: false