patrickvonplaten commited on
Commit
3ee4870
1 Parent(s): a7acc83

new images

Browse files
generated_image.png DELETED
Binary file (177 kB)
 
generated_image_pipeline.png → generated_image_0.png RENAMED
File without changes
generated_image_1.png ADDED
generated_image_2.png ADDED
generated_image_3.png ADDED
generated_image_unrolled.png DELETED
Binary file (119 kB)
 
run.py CHANGED
@@ -5,6 +5,8 @@ import PIL.Image
5
  import numpy as np
6
  import tqdm
7
 
 
 
8
  # 1. Unroll the full loop
9
  # ==================================================================
10
  # load all models
@@ -19,7 +21,7 @@ unet.to(torch_device)
19
  vqvae.to(torch_device)
20
 
21
  # generate gaussian noise to be decoded
22
- generator = torch.manual_seed(0)
23
  noise = torch.randn(
24
  (1, unet.in_channels, unet.image_size, unet.image_size),
25
  generator=generator,
@@ -50,8 +52,6 @@ image_processed = (image_processed + 1.0) * 127.5
50
  image_processed = image_processed.numpy().astype(np.uint8)
51
  image_pil = PIL.Image.fromarray(image_processed[0])
52
 
53
- image_pil.save("generated_image_unrolled.png")
54
-
55
 
56
  # 2. Use pipeline
57
  # ==================================================================
@@ -64,7 +64,7 @@ import tqdm
64
  pipeline = LatentDiffusionUncondPipeline.from_pretrained("./")
65
 
66
  # generatae image by calling the pipeline
67
- generator = torch.manual_seed(0)
68
  image = pipeline(generator=generator, num_inference_steps=200)["sample"]
69
 
70
  # process image
@@ -73,4 +73,4 @@ image_processed = (image_processed + 1.0) * 127.5
73
  image_processed = image_processed.numpy().astype(np.uint8)
74
  image_pil = PIL.Image.fromarray(image_processed[0])
75
 
76
- image_pil.save("generated_image_pipeline.png")
 
5
  import numpy as np
6
  import tqdm
7
 
8
+ seed = 3
9
+
10
  # 1. Unroll the full loop
11
  # ==================================================================
12
  # load all models
 
21
  vqvae.to(torch_device)
22
 
23
  # generate gaussian noise to be decoded
24
+ generator = torch.manual_seed(seed)
25
  noise = torch.randn(
26
  (1, unet.in_channels, unet.image_size, unet.image_size),
27
  generator=generator,
 
52
  image_processed = image_processed.numpy().astype(np.uint8)
53
  image_pil = PIL.Image.fromarray(image_processed[0])
54
 
 
 
55
 
56
  # 2. Use pipeline
57
  # ==================================================================
 
64
  pipeline = LatentDiffusionUncondPipeline.from_pretrained("./")
65
 
66
  # generatae image by calling the pipeline
67
+ generator = torch.manual_seed(seed)
68
  image = pipeline(generator=generator, num_inference_steps=200)["sample"]
69
 
70
  # process image
 
73
  image_processed = image_processed.numpy().astype(np.uint8)
74
  image_pil = PIL.Image.fromarray(image_processed[0])
75
 
76
+ image_pil.save(f"generated_image_{seed}.png")