Faceswap Forum

Model：Dlight
Batch Size：4
Iterations: 520728

Will it be clearer if i continue training?

config": {
"coverage": 80,
"optimizer": "adam",
"learning_rate": 5e-05,
"allow_growth": false,
"mixed_precision": false,
"convert_batchsize": 16,
"loss_function": "ssim",
"mask_loss_function": "mse",
"l2_reg_term": 100,
"eye_multiplier": 12,
"mouth_multiplier": 8,
"penalized_mask_loss": true,
"mask_type": "extended",
"mask_blur_kernel": 3,
"mask_threshold": 4,
"learn_mask": false,
"features": "best",
"details": "good",
"output_size": 384

At this point it's entirely subjective. Honestly I think your swap already looks pretty good and would try a test convert to see if the result was good enough in motion.

Thank you very much.
I am a C# software engineer and just started learning this project.
Thank you very much for your answer.

Convert Result:

Video : 1920X1072

Do you think continued training can improve the face picture quality（output size 384px）? For example, facial details, skin texture?

The same scene, not sure it is the same frame

The same scene, not sure it is the same frame