Hi and first of all thanks for your great work!
I am a beginner in this field. I encountered a problem when running the code. When I open the feature bank, and appearance_dim=32 the resulting output are worse than when not using them. Why is that? data is Mip-NeRF bicycle.
INFO: args: Namespace(add_color_dist=False, add_cov_dist=False, add_opacity_dist=False, appearance_dim=32, appearance_lr_delay_mult=0.01, appearance_lr_final=0.0005, appearance_lr_init=0.05, appearance_lr_max_steps=30000, checkpoint_iterations=[], compute_cov3D_python=False, convert_SHs_python=False, data_device='cuda', debug=False, debug_from=-1, densify_grad_threshold=0.0002, detect_anomaly=False, ds=1, eval=True, feat_dim=32, feature_lr=0.0075, gpu='-1', images='images', ip='127.0.0.1', iterations=30000, lambda_dssim=0.2, lod=0, lowpoly=False, min_opacity=0.005, mlp_color_lr_delay_mult=0.01, mlp_color_lr_final=5e-05, mlp_color_lr_init=0.008, mlp_color_lr_max_steps=30000, mlp_cov_lr_delay_mult=0.01, mlp_cov_lr_final=0.004, mlp_cov_lr_init=0.004, mlp_cov_lr_max_steps=30000, mlp_feat_fusion_lr_delay_mult=0.01, mlp_feat_fusion_lr_final=5e-05, mlp_feat_fusion_lr_init=0.08, mlp_feat_fusion_lr_max_steps=30000, mlp_feat_multi_lr_delay_mult=0.01, mlp_feat_multi_lr_final=5e-05, mlp_feat_multi_lr_init=0.08, mlp_feat_multi_lr_max_steps=30000, mlp_featurebank_lr_delay_mult=0.01, mlp_featurebank_lr_final=1e-05, mlp_featurebank_lr_init=0.01, mlp_featurebank_lr_max_steps=30000, mlp_fuse_lr_delay_mult=0.01, mlp_fuse_lr_final=5e-05, mlp_fuse_lr_init=0.08, mlp_fuse_lr_max_steps=30000, mlp_opacity_lr_delay_mult=0.01, mlp_opacity_lr_final=2e-05, mlp_opacity_lr_init=0.002, mlp_opacity_lr_max_steps=30000, mlp_pos_lr_delay_mult=0.01, mlp_pos_lr_final=5e-05, mlp_pos_lr_init=0.08, mlp_pos_lr_max_steps=30000, model_path='outputs/Mip-NeRF/bicycle/baseline/2025-08-14_06:33:02', n_offsets=10, offset_lr_delay_mult=0.01, offset_lr_final=0.0001, offset_lr_init=0.01, offset_lr_max_steps=30000, opacity_lr=0.02, percent_dense=0.01, port=21463, position_lr_delay_mult=0.01, position_lr_final=0.0, position_lr_init=0.0, position_lr_max_steps=30000, quiet=False, ratio=1, resolution=-1, rotation_lr=0.002, save_iterations=[30000, 30000], scaling_lr=0.007, sh_degree=3, source_path='data/Mip-NeRF/bicycle', start_checkpoint=None, start_stat=500, success_threshold=0.8, test_iterations=[30000], undistorted=False, update_depth=3, update_from=1500, update_hierachy_factor=4, update_init_factor=16, update_interval=100, update_until=15000, use_feat_bank=True, use_wandb=False, voxel_size=0.001, warmup=False, white_background=False)..

self.use_feat_bank = True and appearance_dim=32

self.use_feat_bank = False and appearance_dim=0