blakeyy / relational-rfdnet Goto Github PK

Relational-RfD-Net: A Semantic Instance Reconstruction framework using Attention
[Ingo Blakowski], [Trung Quoc Nguyen]

Ground-truth	Prediction (RfD)	Prediction (Relational-RfD)

To set up the project please refer to the README_original.md of the original RfDNet in this project folder. There you can also see the basic commands to train and test the models.

To control the training and testing the configuration files (see 'configs/config_files/****.yaml') are used.

1. Set before proposal generation MLP (before_prop_gen: True)

2. Or/and set after proposal generation MLP (after_prop_gen: False):

   self_attention:
    appearance_feature_dim: 128
    before_prop_gen: True
    after_prop_gen: False
   

1. Use GT (use_gt_boxsize: True) or predicted (use_gt_boxsize: False) box size.
2. Compute either two box losses before and after the relation module (compute_two_losses: True) or only one box loss after the relation module (compute_two_losses: False).

   relation_module:
    use_relation: False
    method: RelationalProposalModule
    loss: Null
    use_gt_boxsize: True
    compute_two_losses: False
    #use_learned_pos_embed: False
    n_relations: 8 #4
    appearance_feature_dim: 768 #384
    key_feature_dim: 96 
    geo_feature_dim: 96
    isDuplication: False