数字人SyncTalk解压即用N卡整合版_240607

woshidageda

官方演示的模型效果也不行，没musetalk效果好

Danielhan

前来报道

xiaofei450

这个用推理官方的May模型怎么整个脸都是黑的？

xiaofei450

训练的时候报这个是啥情况啊：
Loading model from: C:\code\SyncTalk_240528\python\lib\site-packages\lpips\weights\v0.1\alex.pth
[INFO] Trainer: ngp | 2024-06-18_23-34-29 | cuda | fp16 | model/trial_may
[INFO] #parameters: 768165
[INFO] Loading latest checkpoint ...
[INFO] Latest checkpoint is model/trial_may\checkpoints\ngp_ep0011.pth
[INFO] loaded model.
[INFO] load at epoch 11, global step 60720
[INFO] loaded optimizer.
[INFO] loaded scheduler.
[INFO] loaded scaler.
[INFO] load 100 val frames.
[INFO] load  aud_features: torch.Size([6074, 1, 512])
Loading val data: 100%|█████████████████████████████████████████████████████████████| 100/100 [00:00<00:00, 962.09it/s]
[INFO] eye_area: 0.002303696470335126 - 0.6881041526794434
[INFO] max_epoch = 11
[INFO] load 553 test frames.
[INFO] load  aud_features: torch.Size([6074, 1, 512])
Loading test data: 100%|███████████████████████████████████████████████████████████| 553/553 [00:00<00:00, 1133.35it/s]
[INFO] eye_area: -0.0057049887254834175 - 0.7780876755714417
++> Evaluate at epoch 11 ...
  0% 0/553 [00:00<?, ?it/s]C:\code\SyncTalk_240528\nerf_triplane\utils.py:580: RuntimeWarning: invalid value encountered in cast
  inp = (inp * 255).astype(np.uint8)
C:\code\SyncTalk_240528\python\lib\site-packages\face_alignment\api.py:147: UserWarning: No faces were detected.
  warnings.warn("No faces were detected.")
Traceback (most recent call last):
  File "C:\code\SyncTalk_240528\main.py", line 259, in <module>
    trainer.evaluate(test_loader) # blender has gt, so evaluate it.
  File "C:\code\SyncTalk_240528\nerf_triplane\utils.py", line 1039, in evaluate
    self.evaluate_one_epoch(loader, name)
  File "C:\code\SyncTalk_240528\nerf_triplane\utils.py", line 1403, in evaluate_one_epoch
    metric.update(preds, truths)
  File "C:\code\SyncTalk_240528\nerf_triplane\utils.py", line 589, in update
    lms_pred = self.get_landmarks(preds)
  File "C:\code\SyncTalk_240528\nerf_triplane\utils.py", line 560, in get_landmarks
    lms = self.predictor.get_landmarks(img)[-1]
TypeError: 'NoneType' object is not subscriptable

过去27996789

xiaofei450 发表于 2024-6-18 23:52
训练的时候报这个是啥情况啊：
Loading model from: C:\code\SyncTalk_240528\python\lib\site-packages\lp ...

学习XIA看看能不能学会

woshidageda

下载这个整合包测试了一下,  里边的 bs.npy生成不了, 还需要优化,一下  要不然别人下载也无法用,  或者弄个说明

showmezhen

谢谢分享，等待中~~~

用我一辈子

问下楼主，这个软件只能做512×512分辨率的视频吗，这么小的视频很难商用吧？

登山客

感谢分享

xiao-t

999 零食需要充值多少，给到我？（我没有找到换算的地方）