Spaces:
Sleeping
Sleeping
add requirements and app.py
Browse files- app.py +87 -0
- requirements.txt +4 -0
app.py
ADDED
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pickle
|
2 |
+
import os
|
3 |
+
from sklearn.neighbors import NearestNeighbors
|
4 |
+
import numpy as np
|
5 |
+
num_nn = 20
|
6 |
+
import gradio as gr
|
7 |
+
from PIL import Image
|
8 |
+
|
9 |
+
data_root = '/dccstor/elishc1/datasets/DomainNet'
|
10 |
+
feat_dir = 'brad_feats'
|
11 |
+
domains = ['real', 'painting', 'clipart', 'sketch']
|
12 |
+
shots = '-1'
|
13 |
+
search_domain = 'all'
|
14 |
+
num_results_per_domain = 5
|
15 |
+
src_data_dict = {}
|
16 |
+
if search_domain == 'all':
|
17 |
+
for d in domains:
|
18 |
+
with open(os.path.join(feat_dir, f'dst_{d}_{shots}.pkl'), 'rb') as fp:
|
19 |
+
src_data = pickle.load(fp)
|
20 |
+
|
21 |
+
src_nn_fit = NearestNeighbors(n_neighbors=num_results_per_domain,
|
22 |
+
algorithm='auto', n_jobs=-1).fit(src_data[1])
|
23 |
+
src_data_dict[d] = (src_data,src_nn_fit)
|
24 |
+
else:
|
25 |
+
|
26 |
+
with open(os.path.join(feat_dir, f'dst_{search_domain}_{shots}.pkl'), 'rb') as
|
27 |
+
fp:
|
28 |
+
src_data = pickle.load(fp)
|
29 |
+
src_nn_fit = NearestNeighbors(n_neighbors=num_results_per_domain,
|
30 |
+
algorithm='auto', n_jobs=-1).fit(src_data[1])
|
31 |
+
src_data_dict[search_domain] = (src_data,src_nn_fit)
|
32 |
+
|
33 |
+
dst_data_dict = {}
|
34 |
+
for d in domains:
|
35 |
+
with open(os.path.join(feat_dir, f'src_{d}_{shots}.pkl'), 'rb') as fp:
|
36 |
+
dst_data_dict[d] = pickle.load(fp)
|
37 |
+
|
38 |
+
def query(query_index, query_domain):
|
39 |
+
dst_data = dst_data_dict[query_domain]
|
40 |
+
dst_img_path = os.path.join(data_root, dst_data[0][query_index])
|
41 |
+
img_paths = [dst_img_path]
|
42 |
+
q_cl = dst_img_path.split('/')[-2]
|
43 |
+
captions = [f'Query: {q_cl}']
|
44 |
+
for s_domain, s_data in src_data_dict.items():
|
45 |
+
_, top_n_matches_ids =
|
46 |
+
s_data[1].kneighbors(dst_data[1][query_index:query_index+1])
|
47 |
+
top_n_labels = s_data[0][2][top_n_matches_ids][0]
|
48 |
+
src_img_pths = [os.path.join(data_root, s_data[0][0][ix]) for ix in
|
49 |
+
top_n_matches_ids[0]]
|
50 |
+
img_paths += src_img_pths
|
51 |
+
|
52 |
+
for p in src_img_pths:
|
53 |
+
src_cl = p.split('/')[-2]
|
54 |
+
src_file = p.split('/')[-1]
|
55 |
+
captions.append(src_cl)
|
56 |
+
return tuple([Image.open(p) for p in img_paths])+ tuple(captions)
|
57 |
+
try:
|
58 |
+
demo.close()
|
59 |
+
except:
|
60 |
+
pass
|
61 |
+
demo = gr.Blocks()
|
62 |
+
with demo:
|
63 |
+
gr.Markdown('## Select Query Domain: ')
|
64 |
+
domain_drop = gr.Dropdown(domains)
|
65 |
+
# domain_select_button = gr.Button("Select Domain")
|
66 |
+
slider = gr.Slider(0, 1000)
|
67 |
+
image_button = gr.Button("Run")
|
68 |
+
|
69 |
+
gr.Markdown('# Query Image')
|
70 |
+
src_cap = gr.Label()
|
71 |
+
src_img = gr.Image()
|
72 |
+
|
73 |
+
|
74 |
+
out_images = []
|
75 |
+
out_captions = []
|
76 |
+
for d in domains:
|
77 |
+
gr.Markdown(f'# {d.title()} Domain Images')
|
78 |
+
with gr.Row():
|
79 |
+
for _ in range(num_results_per_domain):
|
80 |
+
with gr.Column():
|
81 |
+
out_captions.append(gr.Label())
|
82 |
+
out_images.append(gr.Image())
|
83 |
+
|
84 |
+
image_button.click(query, inputs=[slider, domain_drop],
|
85 |
+
outputs=[src_img]+out_images +[src_cap]+ out_captions)
|
86 |
+
demo.launch(share=True)
|
87 |
+
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
numpy==1.20.1
|
2 |
+
Pillow==8.2.0
|
3 |
+
scikit-learn==0.24.1
|
4 |
+
|