Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -66,7 +66,10 @@ is_cuda = torch.cuda.is_available()
|
|
66 |
device = torch.device("cuda" if is_cuda else "cpu")
|
67 |
print(device)
|
68 |
clip_model, clip_preprocess = clip.load("ViT-B/32", device=device, jit=False, download_root='./') # Must set jit=False for training
|
69 |
-
|
|
|
|
|
|
|
70 |
clip_model.eval()
|
71 |
for p in clip_model.parameters():
|
72 |
p.requires_grad = False
|
@@ -101,15 +104,16 @@ print('loading transformer checkpoint from {}'.format(args.resume_trans))
|
|
101 |
ckpt = torch.load(args.resume_trans, map_location='cpu')
|
102 |
trans_encoder.load_state_dict(ckpt['trans'], strict=True)
|
103 |
trans_encoder.eval()
|
|
|
104 |
mean = torch.from_numpy(np.load('./checkpoints/t2m/VQVAEV3_CB1024_CMT_H1024_NRES3/meta/mean.npy'))
|
105 |
std = torch.from_numpy(np.load('./checkpoints/t2m/VQVAEV3_CB1024_CMT_H1024_NRES3/meta/std.npy'))
|
|
|
106 |
if is_cuda:
|
107 |
net.cuda()
|
108 |
trans_encoder.cuda()
|
109 |
mean = mean.cuda()
|
110 |
std = std.cuda()
|
111 |
|
112 |
-
|
113 |
def render(motions, device_id=0, name='test_vis'):
|
114 |
frames, njoints, nfeats = motions.shape
|
115 |
MINS = motions.min(axis=0).min(axis=0)
|
|
|
66 |
device = torch.device("cuda" if is_cuda else "cpu")
|
67 |
print(device)
|
68 |
clip_model, clip_preprocess = clip.load("ViT-B/32", device=device, jit=False, download_root='./') # Must set jit=False for training
|
69 |
+
|
70 |
+
if is_cuda:
|
71 |
+
clip.model.convert_weights(clip_model)
|
72 |
+
|
73 |
clip_model.eval()
|
74 |
for p in clip_model.parameters():
|
75 |
p.requires_grad = False
|
|
|
104 |
ckpt = torch.load(args.resume_trans, map_location='cpu')
|
105 |
trans_encoder.load_state_dict(ckpt['trans'], strict=True)
|
106 |
trans_encoder.eval()
|
107 |
+
|
108 |
mean = torch.from_numpy(np.load('./checkpoints/t2m/VQVAEV3_CB1024_CMT_H1024_NRES3/meta/mean.npy'))
|
109 |
std = torch.from_numpy(np.load('./checkpoints/t2m/VQVAEV3_CB1024_CMT_H1024_NRES3/meta/std.npy'))
|
110 |
+
|
111 |
if is_cuda:
|
112 |
net.cuda()
|
113 |
trans_encoder.cuda()
|
114 |
mean = mean.cuda()
|
115 |
std = std.cuda()
|
116 |
|
|
|
117 |
def render(motions, device_id=0, name='test_vis'):
|
118 |
frames, njoints, nfeats = motions.shape
|
119 |
MINS = motions.min(axis=0).min(axis=0)
|