Skip to content
This repository has been archived by the owner on Nov 9, 2023. It is now read-only.

Commit

Permalink
update training instruction
Browse files Browse the repository at this point in the history
  • Loading branch information
@iperov
iperov committed Aug 17, 2022
1 parent 0b26ba2 commit 09ef6d3
Showing 1 changed file with 15 additions and 15 deletions.
30 changes: 15 additions & 15 deletions doc/user_faq/user_faq.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,33 +59,33 @@ Gather 5000+ samples of your face with various conditions using webcam which wil

Here public storage https://disk.yandex.ru/d/7i5XTKIKVg5UUg with facesets and models.

> Using pretrained "RTT model 224.zip" from public storage (see above)
> Using pretrained "RTT model 224 V2.zip" from public storage (see above)
Make a backup before every stage !

1. place RTM WF Faceset from public storage (see above) to workspace/data_dst/aligned
1. place RTM WF Faceset V2 from public storage (see above) to workspace/data_dst/aligned

2. place your celeb to workspace/data_src/aligned

3. do not change settings. Train +500.000

4. replace dst faceset with your faceset in workspace/data_dst/aligned

5. continue train +500.000
5. continue train +500.000, (optional) deleting inter_AB.npy every 100.000 (save, delete, continue run)

6. random_warp:OFF, train +500.000

7. GAN 0.1 power, patch size 28, gan_dims:32. Train until the src loss value has not increased in the last 12 hours.

8. finalize model by disabling masked training for 100-200 (not thousand) iterations.
8. (optional) finalize model by disabling masked training for 100-200 (not thousand) iterations.

> Using SAEHD model from scratch.
res:224, WF, archi:liae-udt, ae_dims:512, e_dims:64, d_dims:64, d_mask_dims:32, eyes_mouth_prio:Y, blur_out_mask:Y, uniform_yaw:Y, lr_dropout:Y, batch:8. Others by default.
res:224, WF, archi:liae-udt, ae_dims:512, e_dims:64, d_dims:64, d_mask_dims:32, eyes_mouth_prio:N, blur_out_mask:Y, uniform_yaw:Y, lr_dropout:Y, batch:8. Others by default.

Make a backup before every stage !

1. place RTM WF Faceset from public storage (see above) to workspace/data_dst/aligned
1. place RTM WF Faceset V2 from public storage (see above) to workspace/data_dst/aligned

2. place your celeb to workspace/data_src/aligned

Expand All @@ -101,7 +101,7 @@ Make a backup before every stage !

8. GAN 0.1 power, gan_dims:32, Train until the src loss value has not increased in the last 12 hours.

9. finalize model by disabling masked training for 100-200 (not thousand) iterations.
9. (optional) finalize model by disabling masked training for 100-200 (not thousand) iterations.

10. export the model in .dfm format for use in DeepFaceLive. You can also try ordering a deepfake model from someone in Discord or forum.

Expand All @@ -119,31 +119,31 @@ Src faceset is celebrity. Must be diverse enough in yaw, light and shadow condit
Do not mix different age. The best result is obtained when the face is filmed from a short period of time and does not change the makeup and structure.
Src faceset should be xseg'ed and applied. You can apply Generic XSeg to src faceset.

> Using pretrained "RTT model 224.zip" from public storage (see above)
> Using pretrained "RTT model 224 V2.zip" from public storage (see above)
Make a backup before every stage !

1. place RTM WF Faceset from public storage (see above) to workspace/data_dst/aligned
1. place RTM WF Faceset V2 from public storage (see above) to workspace/data_dst/aligned

2. place your celeb to workspace/data_src/aligned

3. place model folder to workspace/model

4. do not change settings, train +500.000 iters
4. do not change settings, train +500.000 iters, + (optional) deleting inter_AB.npy every 100.000 (save, delete, continue run)

5. random_warp OFF, train +500.000, periodically (every 100.000 iters) disable masked training for 5.000 iters and enable again
5. random_warp OFF, train +500.000, + (optional) periodically (every 100.000 iters) disable masked training for 5.000 iters and enable again

6. GAN 0.1 power, patch size 28, gan_dims:32. Train until the src loss value has not increased in the last 12 hours.

7. finalize model by disabling masked training for 100-200 (not thousand) iterations.
7. (optional) finalize model by disabling masked training for 100-200 (not thousand) iterations.

> Using SAEHD model from scratch
res:224, WF, archi:liae-udt, ae_dims:512, e_dims:64, d_dims:64, d_mask_dims:32, eyes_mouth_prio:Y, blur_out_mask:Y, uniform_yaw:Y, lr_dropout:Y, batch:8. Others by default.
res:224, WF, archi:liae-udt, ae_dims:512, e_dims:64, d_dims:64, d_mask_dims:32, eyes_mouth_prio:N, blur_out_mask:Y, uniform_yaw:Y, lr_dropout:Y, batch:8. Others by default.

Make a backup before every stage !

1. place RTM WF Faceset from public storage (see above) to workspace/data_dst/aligned
1. place RTM WF Faceset V2 from public storage (see above) to workspace/data_dst/aligned

2. place your celeb to workspace/data_src/aligned

Expand All @@ -155,7 +155,7 @@ Make a backup before every stage !

6. GAN 0.1 power, gan_dims:32. Train until the src loss value has not increased in the last 12 hours.

7. finalize model by disabling masked training for 100-200 (not thousand) iterations.
7. (optional) finalize model by disabling masked training for 100-200 (not thousand) iterations.

> reusing trained SAEHD RTM model
Expand Down

0 comments on commit 09ef6d3

Please sign in to comment.