johndpope MegaPortrait-hack issues

johndpope / MegaPortrait-hack

Using Claude Opus to reverse engineer code from MegaPortraits: One-shot Megapixel Neural Head Avatars

https://arxiv.org/abs/2207.07621

82 stars 8 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

ModNet - completely missed this

#59 johndpope opened 2 weeks ago
0
I would like to ask if it’s realistic for training one epoch on a single video to take around 3 minutes. Is this accurate?

#58 zzyfight0703 opened 2 weeks ago
1
Demo video and evaluation

#57 fkcptlst opened 2 months ago
1
Training on CelebV-HQ

#56 kuwan2e opened 2 months ago
1
Feat/warp fix

#55 johndpope closed 2 weeks ago
1
Question on your warping implementation

#54 yataoz opened 3 months ago
3
Inter-Reconstructed Feature disentanglement(IRFD) - cheaper / faster / cleaner disentanglement - https://arxiv.org/pdf/2405.07257

#53 johndpope closed 3 months ago
2
Are you still bottlenecked on GPU access?

#52 varunjain99 opened 5 months ago
1
dpe from VASA

#51 johndpope closed 5 months ago
0
Are high resolution and student models trainable now?

#50 vishwaraghava009 closed 5 months ago
3
replace preprocessing code

#49 johndpope closed 2 months ago
0
How is the result now?

#48 JZArray closed 2 months ago
2
Feat/celebvhq2 - MultiScalePatchDiscriminator

#47 johndpope closed 2 weeks ago
1
replace patchgan -> multiscale patchgan - should improve image quality

#46 johndpope closed 5 months ago
1
improve tensorboard loss graphs.

#45 johndpope closed 3 months ago
1
cosface loss bug

#44 hazard-10 closed 5 months ago
3
Feat/sharpen image

#43 johndpope closed 5 months ago
0
synthesize cross-identity image pairs for the training of our control module

#42 johndpope closed 2 months ago
0
subsampling on perceptual loss trick from fsrt paper

#41 johndpope closed 2 months ago
0
Update Dataloader to use voxceleb2 / as well / instead of celebvhq

#40 johndpope closed 2 months ago
1
Audit Flops - generator / discriminator

#39 johndpope closed 2 months ago
0
Design for preprocessor at scale / distributed

#38 johndpope closed 2 months ago
2
Feat/celeb v hq

#37 johndpope closed 5 months ago
0
Current results of training - epoch 4

#36 johndpope closed 2 months ago
56
multi-scale patch discriminator + loss

#35 johndpope closed 5 months ago
2
preprocessing steps - warp / crop / align / mask

#34 johndpope closed 5 months ago
2
DRAFT - Feat/32 cycle consistency

#33 johndpope closed 5 months ago
3
cycle consistency loss - produce cross-reenacted image: ˆx𝑠∗→𝑑 = Gbase (x𝑠∗ , x𝑑 )

#32 johndpope closed 5 months ago
2
confusoin about renset50 encoder for global descriptor es

#31 hazard-10 closed 5 months ago
5
experiment on reconstructing the input image

#30 flyingshan closed 5 months ago
4
About GPU info

#29 samsara-ku closed 5 months ago
5
Feat/16 make faster

#28 johndpope closed 5 months ago
1
loss target

#27 ChenyangWang95 closed 5 months ago
7
Reachitect base model to make faster training / use less ram

#26 johndpope closed 2 months ago
8
Implement pre-trained gaze estimation model

#25 johndpope opened 5 months ago
3
Revert "change some model configs"

#24 johndpope closed 6 months ago
0
PR key: zs&es vectors, change model outputs and structure

#23 JackAILab closed 5 months ago
7
change some model configs

#22 JackAILab closed 6 months ago
3
DRAFT - training overhaul - wip

#21 johndpope closed 5 months ago
1
Feat/19 sixdrepnet

#20 johndpope closed 6 months ago
1
HeadPose net - with resnet18 backbone / pitch / yaw / roll

#19 johndpope closed 5 months ago
13
Impossible G3D architecture

#18 robinchm closed 6 months ago
4
Warping twice?

#17 robinchm closed 6 months ago
1
Why not use the newer & improved MetaPortrait instead of reimplementing MegaPortrait?

#16 mrT333 closed 5 months ago
3
WarpingNet / flow fields / grids - reference

#15 johndpope closed 5 months ago
6
Stage 1 - training losses - todo contrastive_loss

#14 johndpope closed 5 months ago
5
fyi - custom resnet50 - reference code fc / maxpool

#13 johndpope closed 5 months ago
1
Stage 1 - apply_warping_field - vs seems to be wrong size

#12 johndpope closed 6 months ago
2
Stage1 training- warpgenerator blows up OOM

#11 johndpope closed 6 months ago
1
es global descriptor - how to plug this back into Gbase

#10 johndpope closed 6 months ago
0