Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -24,8 +24,8 @@ for idx, url in enumerate(urls):
|
|
24 |
image.save(f"image_{idx}.png")
|
25 |
|
26 |
def process_image(image):
|
27 |
-
# prepare
|
28 |
-
batch_size =
|
29 |
encoding = feature_extractor([image for _ in range(batch_size)], return_tensors="pt")
|
30 |
|
31 |
# create primers
|
@@ -35,9 +35,9 @@ def process_image(image):
|
|
35 |
n_px_crop = 16
|
36 |
primers = samples.reshape(-1,n_px*n_px)[:,:n_px_crop*n_px] # crop top n_px_crop rows. These will be the conditioning tokens
|
37 |
|
38 |
-
# get conditioned image (from first primer tensor)
|
39 |
primers_img = np.reshape(np.rint(127.5 * (clusters[primers[0]] + 1.0)), [n_px_crop,n_px, 3]).astype(np.uint8)
|
40 |
-
primers_img =
|
41 |
|
42 |
# generate (no beam search)
|
43 |
context = np.concatenate((np.full((batch_size, 1), model.config.vocab_size - 1), primers), axis=1)
|
@@ -48,6 +48,8 @@ def process_image(image):
|
|
48 |
samples = output[:,1:].cpu().detach().numpy()
|
49 |
samples_img = [np.reshape(np.rint(127.5 * (clusters[s] + 1.0)), [n_px, n_px, 3]).astype(np.uint8) for s in samples]
|
50 |
|
|
|
|
|
51 |
# stack images horizontally
|
52 |
row1 = np.hstack(samples_img[:4])
|
53 |
row2 = np.hstack(samples_img[4:])
|
@@ -56,17 +58,16 @@ def process_image(image):
|
|
56 |
# return as PIL Image
|
57 |
completion = Image.fromarray(result)
|
58 |
|
59 |
-
return
|
60 |
|
61 |
title = "Interactive demo: ImageGPT"
|
62 |
description = "Demo for OpenAI's ImageGPT: Generative Pretraining from Pixels. To use it, simply upload an image or use the example image below and click 'submit'. Results will show up in a few seconds."
|
63 |
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2109.10282'>ImageGPT: Generative Pretraining from Pixels</a> | <a href='https://openai.com/blog/image-gpt/'>Official blog</a></p>"
|
64 |
examples =[f"image_{idx}.png" for idx in range(len(urls))]
|
65 |
|
66 |
-
labels = ["Conditioned image:", "Completions:"]
|
67 |
iface = gr.Interface(fn=process_image,
|
68 |
inputs=gr.inputs.Image(type="pil"),
|
69 |
-
outputs=
|
70 |
title=title,
|
71 |
description=description,
|
72 |
article=article,
|
|
|
24 |
image.save(f"image_{idx}.png")
|
25 |
|
26 |
def process_image(image):
|
27 |
+
# prepare 7 images, shape (7, 1024)
|
28 |
+
batch_size = 7
|
29 |
encoding = feature_extractor([image for _ in range(batch_size)], return_tensors="pt")
|
30 |
|
31 |
# create primers
|
|
|
35 |
n_px_crop = 16
|
36 |
primers = samples.reshape(-1,n_px*n_px)[:,:n_px_crop*n_px] # crop top n_px_crop rows. These will be the conditioning tokens
|
37 |
|
38 |
+
# get conditioned image (from first primer tensor), padded with black pixels to be 32x32
|
39 |
primers_img = np.reshape(np.rint(127.5 * (clusters[primers[0]] + 1.0)), [n_px_crop,n_px, 3]).astype(np.uint8)
|
40 |
+
primers_img = np.pad(primers_img, pad_width=((0,16), (0,0), (0,0)), mode="constant")
|
41 |
|
42 |
# generate (no beam search)
|
43 |
context = np.concatenate((np.full((batch_size, 1), model.config.vocab_size - 1), primers), axis=1)
|
|
|
48 |
samples = output[:,1:].cpu().detach().numpy()
|
49 |
samples_img = [np.reshape(np.rint(127.5 * (clusters[s] + 1.0)), [n_px, n_px, 3]).astype(np.uint8) for s in samples]
|
50 |
|
51 |
+
samples_img = [primers_img] + samples_img
|
52 |
+
|
53 |
# stack images horizontally
|
54 |
row1 = np.hstack(samples_img[:4])
|
55 |
row2 = np.hstack(samples_img[4:])
|
|
|
58 |
# return as PIL Image
|
59 |
completion = Image.fromarray(result)
|
60 |
|
61 |
+
return completion
|
62 |
|
63 |
title = "Interactive demo: ImageGPT"
|
64 |
description = "Demo for OpenAI's ImageGPT: Generative Pretraining from Pixels. To use it, simply upload an image or use the example image below and click 'submit'. Results will show up in a few seconds."
|
65 |
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2109.10282'>ImageGPT: Generative Pretraining from Pixels</a> | <a href='https://openai.com/blog/image-gpt/'>Official blog</a></p>"
|
66 |
examples =[f"image_{idx}.png" for idx in range(len(urls))]
|
67 |
|
|
|
68 |
iface = gr.Interface(fn=process_image,
|
69 |
inputs=gr.inputs.Image(type="pil"),
|
70 |
+
outputs=gr.outputs.Image(type="pil", label="Model input + completions"),
|
71 |
title=title,
|
72 |
description=description,
|
73 |
article=article,
|