Search found 2638 matches

by torzdf
Wed Aug 18, 2021 10:08 am
Forum: Training Discussion
Topic: Around what learning rate and epsilon should it be set for 384 px Phaze-A models using mixed precision?
Replies: 3
Views: 8615

Re: Around what learning rate and epsilon should it be set for 384 px Phaze-A models using mixed precision?

-5 should be ok for mixed precision. As for learning rate, whatever works. There are no hard and fast figures, I'm afraid. If 5e-5 doesn't work, try lowering to 4.5e-5, if that doesn't work, drop again. Etc. etc. Also, make sure your faceswap is updated. I pushed an update at the beginning of august...
by torzdf
Tue Aug 17, 2021 9:30 am
Forum: Extract Support
Topic: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'
Replies: 15
Views: 16774

Re: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'

Essayez d'abord les étapes indiquées dans le message d'erreur d'origine. Faites-moi savoir si cela le résout. Traduction automatique : Vous ne disposez pas de suffisamment de mémoire GPU pour exécuter la détection à la taille de lot sélectionnée. Vous pouvez essayer plusieurs choses: Fermez toute au...
by torzdf
Tue Aug 17, 2021 9:23 am
Forum: Training Discussion
Topic: Strange Red Area/Artifact Creeping into Model
Replies: 26
Views: 24924

Re: Strange Red Area/Artifact Creeping into Model

Mask won't cause that kind of issue. However, I used vgg-clear (bisenet-fp had not yet been implemented). Since then I have trained with bisenet-fp just fine.

by torzdf
Sun Aug 15, 2021 7:05 pm
Forum: Extract Discussion
Topic: Ensuring Both Faces are Extracted From a Frame (Even if they are the same - VR 3D Footage)
Replies: 3
Views: 7694

Re: Ensuring Both Faces are Extracted From a Frame (Even if they are the same - VR 3D Footage)

The extraction process makes no comparison on faces. It will extract all faces it finds in a video. If it's not extracting one of the faces then it's not finding the face.

Unfortunately I have no experience faceswapping with VR, so this is about the extent that I can help, I'm afraid.

by torzdf
Sun Aug 15, 2021 5:32 pm
Forum: Extract Support
Topic: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'
Replies: 15
Views: 16774

Re: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'

Dude. I don't know what I can say. I have given you instructions twice now about what you need to provide to get support and you still haven't provided it.

by torzdf
Sun Aug 15, 2021 10:42 am
Forum: Training Discussion
Topic: Strange Red Area/Artifact Creeping into Model
Replies: 26
Views: 24924

Re: Strange Red Area/Artifact Creeping into Model

Yeah, sorry. That's not an issue and can be ignored. It basically appears because of the way Phaze-A is constructed (multiple encoders) and not wanting to include every possible encoder as a configuration option, hence we store "keras-encoder" in the configuration file, which has to be rew...
by torzdf
Sun Aug 15, 2021 10:31 am
Forum: Extract Discussion
Topic: Could not broadcast input array from shape
Replies: 11
Views: 13888

Re: Could not broadcast input array from shape

Ok, this issue looks specific to 30xx cards.... I am pretty sure the version of Cuda/cuDNN is not correct for the version of Tensorflow you have installed. You'll need to fix that up and see if it works. Ultimately, these kind of mismatches can manifest themselves in strange ways, as you are discove...
by torzdf
Sun Aug 15, 2021 10:01 am
Forum: Training Discussion
Topic: Strange Red Area/Artifact Creeping into Model
Replies: 26
Views: 24924

Re: Strange Red Area/Artifact Creeping into Model

I'm honestly at a loss. It may be that the issue in the model has already occurred prior to 150k, it just does not represent in the loss until later. I would start the model again and pray. Not much help, I know. I have just never hit this kind of issue before. [mention]bryanlyon[/mention] may have ...
by torzdf
Sat Aug 14, 2021 11:18 am
Forum: Training Support
Topic: ValueError: could not broadcast input array from shape (512,512,3) into shape (512,512)
Replies: 3
Views: 8583

Re: ValueError: could not broadcast input array from shape (512,512,3) into shape (512,512)

If I were to guess, I'd say there are black and white images in your training set. This shouldn't matter, and I haven't seen this issue before, but it's worth checking. If you do, remove the black and white images from your training set and see if you can get it running.

by torzdf
Sat Aug 14, 2021 9:40 am
Forum: Training Support
Topic: Model will no longer train after reinstaling faceswap
Replies: 1
Views: 6291

Re: Model will no longer train after reinstaling faceswap

This error happens when a model is trained with tf.keras and you attempt to load it in Keras. As AMD cards do not use tf.Keras, I can only assume that this model was created on an Nvidia card or on CPU and you are now attempting to load it on to an AMD device.

by torzdf
Sat Aug 14, 2021 9:37 am
Forum: Training Discussion
Topic: Strange Red Area/Artifact Creeping into Model
Replies: 26
Views: 24924

Re: Strange Red Area/Artifact Creeping into Model

It may be data driven, may be a combination of images in the batch, may be a power spike... honestly, I really don't know what would have caused the issue. You can probably still rescue the model by rolling back to a snapshot pre "spike" (i.e. back to c. 150k iters) and continuing from the...
by torzdf
Fri Aug 13, 2021 10:58 pm
Forum: Training Discussion
Topic: Strange Red Area/Artifact Creeping into Model
Replies: 26
Views: 24924

Re: Strange Red Area/Artifact Creeping into Model

Jumps like that happen sometimes. As long as preview is still improving and it continues to trend down, I wouldn't worry about it too much. That red area is weird. Not seen it before. It may be related to the jump, it may not. I'm hoping that it doesn't mean your model is dead, but I guess only time...
by torzdf
Fri Aug 13, 2021 10:02 am
Forum: Training Support
Topic: Training Crashes switching from NVidia to AMD
Replies: 2
Views: 7513

Re: Training Crashes switching from NVidia to AMD

Unfortunately it is not possible to go cross manufacturer this way. We need to use a totally different backend to train models on AMD machines, there is literally nothing we can do about this if we want to keep AMD support. There are a couple of potential solutions, neither of which I have tested, s...
by torzdf
Thu Aug 12, 2021 9:00 am
Forum: Installation Support
Topic: Where can I find more information about the different Branches
Replies: 1
Views: 8048

Re: Where can I find more information about the different Branches

Don't switch branch unless you're instructed to would be my suggestion. They are generally older snapshots of the app prior to some big change, so people could easily roll back. The other reason we might want to switch branch is if a new feature is being tested but is not yet rolled into mainline. T...
by torzdf
Thu Aug 12, 2021 8:59 am
Forum: Training Support
Topic: Conv Aware Init: how long should I wait?
Replies: 1
Views: 6558

Re: Conv Aware Init: how long should I wait?

Ok, most likely Conv Aware has finished and it's something else causing the hang. Conv Aware just shows output as it can take a long time.

Please post your system info (i.e. follow these instructions: app.php/rules#rule-4b)

by torzdf
Thu Aug 12, 2021 8:57 am
Forum: Extract Discussion
Topic: Landmark accuracy
Replies: 3
Views: 11341

Re: Landmark accuracy

Yeah, it doesn't deal with obstructions well. Normally copying alignments from previous or next face and then moving the extract box is enough to remedy this.

by torzdf
Thu Aug 12, 2021 8:55 am
Forum: Extract Support
Topic: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'
Replies: 15
Views: 16774

Re: Extracttion Error: Assertion failed fabs(sc) > DBL_EPSILON in function 'cvFindExtrinsicCameraParams2'

You misunderstood. You need to follow the instructions in the above link. I know the contents of the file. I wrote it :/

In the GUI:

Help > Output System Info

Post the output here.