Spaces:
Sleeping
Sleeping
Daniel Varga
commited on
Commit
·
51b0e53
1
Parent(s):
9cdc9a1
switching to annoy
Browse files- app.py +17 -15
- requirements.txt +1 -0
app.py
CHANGED
@@ -7,6 +7,7 @@ import gradio as gr
|
|
7 |
import numpy as np
|
8 |
import torch
|
9 |
import clip
|
|
|
10 |
|
11 |
|
12 |
CONFIG_PATH = "app.ini"
|
@@ -49,12 +50,18 @@ data = pickle.load(open(pickle_filename, "rb"))
|
|
49 |
# but we use float32 in-memory to avoid numerical issues.
|
50 |
# tbh i'm not sure there are any such issues.
|
51 |
embeddings = data["embeddings"].astype(np.float32)
|
52 |
-
|
53 |
-
image_features /= image_features.norm(dim=-1, keepdim=True)
|
54 |
-
|
55 |
|
56 |
n, d = embeddings.shape
|
57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
filenames = data["filenames"]
|
59 |
|
60 |
urls = [base_url + filename for filename in filenames]
|
@@ -67,29 +74,24 @@ def embed_text(text):
|
|
67 |
with torch.no_grad():
|
68 |
text_features = model.encode_text(tokens)
|
69 |
assert text_features.shape == (1, d)
|
|
|
|
|
70 |
return text_features
|
71 |
|
72 |
|
73 |
-
def similarities(text_features, topk=20):
|
74 |
-
text_features /= text_features.norm(dim=-1, keepdim=True)
|
75 |
-
# the softmax rounds up everything to 1, so does not distinguish between good fits.
|
76 |
-
similarity = (100.0 * image_features @ text_features.T) # .softmax(dim=-1)
|
77 |
-
values, indices = similarity[:, 0].topk(topk)
|
78 |
-
return values, indices
|
79 |
-
|
80 |
-
|
81 |
def image_retrieval_from_text(text):
|
82 |
-
|
|
|
83 |
top_urls = np.array(urls)[indices]
|
84 |
-
return top_urls.tolist(), indices
|
85 |
|
86 |
|
87 |
def image_retrieval_from_image(state, selected_locally):
|
88 |
selected = state[int(selected_locally)]
|
89 |
image_vector = image_features[selected][None, :]
|
90 |
-
|
91 |
top_urls = np.array(urls)[indices]
|
92 |
-
return top_urls.tolist(), indices
|
93 |
|
94 |
|
95 |
with gr.Blocks(css="footer {visibility: hidden}") as demo:
|
|
|
7 |
import numpy as np
|
8 |
import torch
|
9 |
import clip
|
10 |
+
import annoy
|
11 |
|
12 |
|
13 |
CONFIG_PATH = "app.ini"
|
|
|
50 |
# but we use float32 in-memory to avoid numerical issues.
|
51 |
# tbh i'm not sure there are any such issues.
|
52 |
embeddings = data["embeddings"].astype(np.float32)
|
53 |
+
embeddings /= np.linalg.norm(embeddings, axis=-1)[:, None]
|
|
|
|
|
54 |
|
55 |
n, d = embeddings.shape
|
56 |
|
57 |
+
print("annoy indexing")
|
58 |
+
annoy_index = annoy.AnnoyIndex(d, 'angular')
|
59 |
+
for i, vec in enumerate(embeddings):
|
60 |
+
annoy_index.add_item(i, vec)
|
61 |
+
annoy_index.build(10)
|
62 |
+
print("done")
|
63 |
+
|
64 |
+
|
65 |
filenames = data["filenames"]
|
66 |
|
67 |
urls = [base_url + filename for filename in filenames]
|
|
|
74 |
with torch.no_grad():
|
75 |
text_features = model.encode_text(tokens)
|
76 |
assert text_features.shape == (1, d)
|
77 |
+
text_features = text_features.numpy()[0]
|
78 |
+
text_features /= np.linalg.norm(text_features)
|
79 |
return text_features
|
80 |
|
81 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
def image_retrieval_from_text(text):
|
83 |
+
text_features = embed_text(text)
|
84 |
+
indices = annoy_index.get_nns_by_vector(text_features, n=20)
|
85 |
top_urls = np.array(urls)[indices]
|
86 |
+
return top_urls.tolist(), indices
|
87 |
|
88 |
|
89 |
def image_retrieval_from_image(state, selected_locally):
|
90 |
selected = state[int(selected_locally)]
|
91 |
image_vector = image_features[selected][None, :]
|
92 |
+
indices = annoy_index.get_nns_by_item(selected, n=20)
|
93 |
top_urls = np.array(urls)[indices]
|
94 |
+
return top_urls.tolist(), indices
|
95 |
|
96 |
|
97 |
with gr.Blocks(css="footer {visibility: hidden}") as demo:
|
requirements.txt
CHANGED
@@ -1 +1,2 @@
|
|
1 |
git+https://github.com/openai/CLIP.git
|
|
|
|
1 |
git+https://github.com/openai/CLIP.git
|
2 |
+
annoy==1.17.2
|