search

sjtuplayer / few-shot-diffusion

[ICCV 2023] Phasic Content Fusing Diffusion Model with Directional Distribution Consistency for Few-Shot Model Adaption
53 stars 2 forks source link

phasic content fusing model #4

Open fikry102 opened 10 months ago

fikry102 commented 10 months ago

After the basic diffusion model is obtained, you can train phasic content fusing module by: python3 train-recon.py --data_path=$path_to_dataset --ckpt=$path_to_basic_diffusion_checkpoint So, we only use source dataset to train the phasic content fusing module? How should it be related to Figure 2 in the paper? image

sjtuplayer commented 10 months ago

Yes, we only use the source-domain dataset to train the phasic content fusing module in step 2 since this step is only used for initializing the module. Figure 2 is the framework of train-whole.py, which trains the diffusion model and the phasic content fusing module at the same time.

invade-art commented 5 months ago

Yes, we only use the source-domain dataset to train the phasic content fusing module in step 2 since this step is only used for initializing the module. Figure 2 is the framework of train-whole.py, which trains the diffusion model and the phasic content fusing module at the same time.

so? why train the phasic content fusing module in step2

sjtuplayer commented 5 months ago

Yes, we only use the source-domain dataset to train the phasic content fusing module in step 2 since this step is only used for initializing the module. Figure 2 is the framework of train-whole.py, which trains the diffusion model and the phasic content fusing module at the same time.

so? why train the phasic content fusing module in step2

the PCF module is used for keeping the identity consistency during domain adaption.