-
Notifications
You must be signed in to change notification settings - Fork 13
/
test_infer.py
44 lines (36 loc) · 1.45 KB
/
test_infer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
from mcquic.modules.generator_3_self_attn import GeneratorV3SelfAttention
from PIL import Image
from mcquic.data.transforms import (
getTrainingPreprocess,
getTrainingTransform,
getEvalTransform,
)
from torchvision.transforms.functional import to_tensor
import torch.nn.functional as F
import torch
def test():
generator = GeneratorV3SelfAttention(
256,
4096,
[16, 8, 8, 8, 8, 4, 4, 4, 4, 2, 2, 2, 2, 1, 1, 1, 1],
denseNorm=False,
qk_norm=True,
norm_eps=1e-5,
loadFrom="./compressor/dynamic_30000.ckpt",
).eval()
ckpt = torch.load("./generation_saved/latest/saved.ckpt", map_location="cpu")
generator.load_state_dict({k.replace("module.", ""): v for k, v in ckpt['trainer']['_model'].items()})
generator = generator.cuda()
print(list(name for name, _ in generator.named_parameters() if "gamma" in name))
with torch.no_grad():
class_label = torch.tensor([79, 264]).cuda()
predictions, restored = generator(None, class_label)
print("****** PRES:", [pre.shape for pre in predictions], "********")
print(restored.shape)
restored_img = ((restored.detach().cpu().permute(0, 2, 3, 1).numpy() + 1) / 2 * 255). astype("uint8")
# print(restored_img[0])
im = Image.fromarray(restored_img[0])
im.save("./restored_img.png")
# print(sum([F.cross_entropy(pre, gt) for (pre, gt) in zip(predictions, codes)]))
if __name__ == "__main__":
test()