Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
import gradio
|
2 |
import os
|
3 |
import numpy as np
|
4 |
import pandas as pd
|
@@ -8,12 +8,17 @@ import torch
|
|
8 |
from transformers import AutoTokenizer, CLIPTextModelWithProjection
|
9 |
|
10 |
|
|
|
|
|
|
|
|
|
|
|
11 |
DATA_PATH = './data'
|
12 |
|
13 |
ft_visual_features_file = DATA_PATH + '/dataset_v1_visual_features_database.npy'
|
14 |
-
|
|
|
15 |
ft_visual_features_database = np.load(ft_visual_features_file)
|
16 |
-
binary_visual_features = np.load(binary_visual_features_file)
|
17 |
|
18 |
database_csv_path = os.path.join(DATA_PATH, 'dataset_v1.csv')
|
19 |
database_df = pd.read_csv(database_csv_path)
|
@@ -38,7 +43,7 @@ class NearestNeighbors:
|
|
38 |
def fit(self, data, o_data=None):
|
39 |
if self.metric == 'cosine':
|
40 |
data = self.normalize(data)
|
41 |
-
self.index = faiss.IndexFlatIP(data.shape[1])
|
42 |
elif self.metric == 'binary':
|
43 |
self.o_data = data if o_data is None else o_data
|
44 |
#assuming data already packed
|
@@ -47,44 +52,37 @@ class NearestNeighbors:
|
|
47 |
|
48 |
def kneighbors(self, q_data):
|
49 |
if self.metric == 'cosine':
|
50 |
-
|
51 |
-
q_data = self.normalize(q_data)
|
52 |
sim, idx = self.index.search(q_data, self.n_neighbors)
|
53 |
else:
|
54 |
if self.metric == 'binary':
|
55 |
-
print('binary search')
|
56 |
bq_data = np.packbits((q_data > 0.0).astype(bool), axis=1)
|
57 |
print(bq_data.shape, self.index.d)
|
58 |
sim, idx = self.index.search(bq_data, max(self.rerank_from, self.n_neighbors))
|
59 |
|
60 |
if self.rerank_from > self.n_neighbors:
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
sim_float[i,:] = sim_float[i,:][sort_idx]
|
67 |
-
idx[i,:] = idx[i,:][sort_idx]
|
68 |
-
sim = sim_float[:,:self.n_neighbors]
|
69 |
-
idx = idx[:,:self.n_neighbors]
|
70 |
|
71 |
return sim, idx
|
72 |
|
73 |
-
|
|
|
|
|
74 |
def search(search_sentence):
|
75 |
-
my_model = CLIPTextModelWithProjection.from_pretrained("Diangle/clip4clip-webvid")
|
76 |
-
tokenizer = AutoTokenizer.from_pretrained("Diangle/clip4clip-webvid")
|
77 |
-
|
78 |
-
|
79 |
inputs = tokenizer(text=search_sentence , return_tensors="pt", padding=True)
|
80 |
|
81 |
-
outputs =
|
82 |
-
|
83 |
-
text_projection =
|
84 |
text_embeds = outputs[1] @ text_projection
|
85 |
final_output = text_embeds[torch.arange(text_embeds.shape[0]), inputs["input_ids"].argmax(dim=-1)]
|
86 |
|
87 |
-
|
88 |
final_output = final_output / final_output.norm(dim=-1, keepdim=True)
|
89 |
final_output = final_output.cpu().detach().numpy()
|
90 |
sequence_output = final_output / np.sum(final_output**2, axis=1, keepdims=True)
|
@@ -94,12 +92,22 @@ def search(search_sentence):
|
|
94 |
sims, idxs = nn_search.kneighbors(sequence_output)
|
95 |
return database_df.iloc[idxs[0]]['contentUrl'].to_list()
|
96 |
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
import os
|
3 |
import numpy as np
|
4 |
import pandas as pd
|
|
|
8 |
from transformers import AutoTokenizer, CLIPTextModelWithProjection
|
9 |
|
10 |
|
11 |
+
TITLE="""<h1 style="font-size: 42px;" align="center">Video Retrieval</h1>"""
|
12 |
+
|
13 |
+
DESCRIPTION="""This is a video retrieval demo using [Diangle/clip4clip-webvid](https://huggingface.co/Diangle/clip4clip-webvid)."""
|
14 |
+
IMAGE='<img src="./Searchium.png"/>'
|
15 |
+
|
16 |
DATA_PATH = './data'
|
17 |
|
18 |
ft_visual_features_file = DATA_PATH + '/dataset_v1_visual_features_database.npy'
|
19 |
+
|
20 |
+
#load database features:
|
21 |
ft_visual_features_database = np.load(ft_visual_features_file)
|
|
|
22 |
|
23 |
database_csv_path = os.path.join(DATA_PATH, 'dataset_v1.csv')
|
24 |
database_df = pd.read_csv(database_csv_path)
|
|
|
43 |
def fit(self, data, o_data=None):
|
44 |
if self.metric == 'cosine':
|
45 |
data = self.normalize(data)
|
46 |
+
self.index = faiss.IndexFlatIP(data.shape[1])
|
47 |
elif self.metric == 'binary':
|
48 |
self.o_data = data if o_data is None else o_data
|
49 |
#assuming data already packed
|
|
|
52 |
|
53 |
def kneighbors(self, q_data):
|
54 |
if self.metric == 'cosine':
|
55 |
+
q_data = self.normalize(q_data)
|
|
|
56 |
sim, idx = self.index.search(q_data, self.n_neighbors)
|
57 |
else:
|
58 |
if self.metric == 'binary':
|
59 |
+
print('binary search: ')
|
60 |
bq_data = np.packbits((q_data > 0.0).astype(bool), axis=1)
|
61 |
print(bq_data.shape, self.index.d)
|
62 |
sim, idx = self.index.search(bq_data, max(self.rerank_from, self.n_neighbors))
|
63 |
|
64 |
if self.rerank_from > self.n_neighbors:
|
65 |
+
rerank_data = self.o_data[idx[0]]
|
66 |
+
rerank_search = NearestNeighbors(n_neighbors=self.n_neighbors, metric='cosine')
|
67 |
+
rerank_search.fit(rerank_data)
|
68 |
+
sim, re_idxs = rerank_search.kneighbors(q_data)
|
69 |
+
idx = [idx[0][re_idxs[0]]]
|
|
|
|
|
|
|
|
|
70 |
|
71 |
return sim, idx
|
72 |
|
73 |
+
model = CLIPTextModelWithProjection.from_pretrained("Diangle/clip4clip-webvid")
|
74 |
+
tokenizer = AutoTokenizer.from_pretrained("Diangle/clip4clip-webvid")
|
75 |
+
|
76 |
def search(search_sentence):
|
|
|
|
|
|
|
|
|
77 |
inputs = tokenizer(text=search_sentence , return_tensors="pt", padding=True)
|
78 |
|
79 |
+
outputs = model(input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], return_dict=False)
|
80 |
+
# Customized projection layer
|
81 |
+
text_projection = model.state_dict()['text_projection.weight']
|
82 |
text_embeds = outputs[1] @ text_projection
|
83 |
final_output = text_embeds[torch.arange(text_embeds.shape[0]), inputs["input_ids"].argmax(dim=-1)]
|
84 |
|
85 |
+
# Normalization
|
86 |
final_output = final_output / final_output.norm(dim=-1, keepdim=True)
|
87 |
final_output = final_output.cpu().detach().numpy()
|
88 |
sequence_output = final_output / np.sum(final_output**2, axis=1, keepdims=True)
|
|
|
92 |
sims, idxs = nn_search.kneighbors(sequence_output)
|
93 |
return database_df.iloc[idxs[0]]['contentUrl'].to_list()
|
94 |
|
95 |
+
|
96 |
+
with gr.Blocks() as demo:
|
97 |
+
gr.HTML(TITLE)
|
98 |
+
gr.Markdown(DESCRIPTION)
|
99 |
+
gr.HTML(IMAGE)
|
100 |
+
gr.Markdown("Retrieval of top 5 videos relevant to the input sentence: ")
|
101 |
+
with gr.Row():
|
102 |
+
with gr.Column():
|
103 |
+
inp = gr.Textbox(placeholder="Write a sentence.")
|
104 |
+
btn = gr.Button(value="Retrieve")
|
105 |
+
ex = [["a woman waving to the camera"],["a basketball player performing a slam dunk"], ["how to bake a chocolate cake"], ["birds fly in the sky"]]
|
106 |
+
gr.Examples(examples=ex,
|
107 |
+
inputs=[inp],
|
108 |
+
)
|
109 |
+
with gr.Column():
|
110 |
+
out = [gr.Video(format='mp4') for _ in range(5)]
|
111 |
+
btn.click(search, inputs=inp, outputs=out)
|
112 |
+
|
113 |
+
demo.launch(debug=True, share=True)
|