aggregate: mean
attn_heads: 4
base_filters: 64
focal_weight: 0.5
fusion_scales: !!python/tuple
- 20
- 40
- 80
input_size: 320
lr: 0.0003
mask_encoder_dims: &id001 !!python/tuple
- 64
- 128
- 256
- 512
max_epochs: 50
model_kind: retrieval
num_classes: 21
prior_channels: *id001
retriever_image_size: 224
top_k: 4
wd: 0.0001