TomatoBoy90

Most of the time is spent in inference.

I use a single GPU (P 100) for inference

> Most of the time is spent on shadow synthesis. In our online system, the shadow is composed of sketch images. To speed up the reading and writing time, we...

> > After many hours, I finally can run the code 2333. Here are some tips to run the code: > > > > 1. the train_MAKEMIX.txt means the names...

(Talking-Face_PC-AVS) root@1949-2d-dd:/aidata/Talking-Face_PC-AVS# bash experiments/demo_vox.sh ----------------- Options --------------- D_input: single VGGFace_pretrain_path: aspect_ratio: 1.0 audio_nc: 256 augment_target: False batchSize: 16 [default: 2] beta1: 0.5 beta2: 0.999 checkpoints_dir: ./checkpoints clip_len: 1 crop: False...

I read your doc,I find that folder will be train_A, train_B when use --label_nc 0. if I use instance maps, need change floder name?

I also got this erro,have you resolve this isssues?

太棒了，我还想着自己用paddle写，没想到你都写好了，还有预训练模型

> Second this! @thaoshibe the README says code will be updated in March, which I assume meant March 2024. This is super cool work, and I'd love to be able...