| data: |
| resolution: 256 |
| n_transducer_points: 512 |
| model: |
| base_width: 16 |
| cond_dim: 128 |
| n_transducer_freqs: 8 |
| dynamic_conv_kernel: 3 |
| cross_attention_heads: 4 |
| cross_attention_levels: |
| - level1 |
| - level2 |
| - level3 |
| - bottleneck |
| cross_attention_bidirectional: false |
| use_film_decoder: false |
| loss: |
| alpha: 5.0 |
| grad_weight: 0.1 |
| focal_weight: 5.0e-05 |
| focal_temperature: 0.03 |
| focal_warmup_off: 0 |
| focal_warmup_epochs: 3 |
| box_weight: 0.0 |
| train: |
| seed: 0 |
| epochs: 15 |
| batch_size: 4 |
| lr: 3.0e-05 |
| weight_decay: 0.0001 |
| grad_clip: 1.0 |
| num_workers: 4 |
| val_every: 1 |
| precision: pure-bf16 |
| grad_checkpoint_encoder: false |
| channels_last: false |
| compile: false |
| save_every_epochs: 1 |
| wandb_project: deeptfus-reproduction |
| wandb_entity: mason-wang |
| eval: |
| voxel_size_mm: 0.5 |
| focal_threshold_db: -6.0 |
| off_target_min_dist_mm: 10.0 |
| n_warmup_inferences: 3 |
| save_predictions: false |
| output: |
| run_dir: runs/deeptfus_ft_E_anchored_aggressive |
|
|
