search

sunshangquan / TransMamba

23 stars 5 forks source link

Configuration parameter #2

Open hhhhhhalf opened 1 month ago

hhhhhhalf commented 1 month ago

Hello, thank you very much for your excellent work. When I read the code, I found some differences from those mentioned in the paper. The following is the model configuration content of the configuration file Deraining_transMamba_Rain200L.yml: network_g: type: TransMamba inp_channels: 3 out_channels: 3 dim: 36 num_blocks: [1,2,3,4] num_refinement_blocks: 2 heads: [1,2,4,8] ffn_expansion_factor: 1.6667 bias: False LayerNorm_type: WithBias dual_pixel_task: False

My puzzle: It says that num_blocks is set to [1, 3, 4, 4] and ffn_expansion_factor is set to 2.667. Or is this a different configuration for different data sets?

sunshangquan commented 1 month ago

Hi @hhhhhhalf , thank you for your interest and attention. Yes, the uploaded weights and settings are for Rain200H/L and are smaller versions. But their results should be fine. I will be updating them in couple of days.

hhhhhhalf commented 1 month ago

OK, I see. Thank you for your reply!