Spaces:
Runtime error
Runtime error
Update app
Browse files
app.py
CHANGED
@@ -148,8 +148,8 @@ def extract_images_and_embeds(video_id, video_path, subtitles, output, expanded=
|
|
148 |
if success:
|
149 |
img_fname = f'{video_id}_{idx:06d}'
|
150 |
img_fpath = os.path.join(output, 'frames', img_fname + '.jpg')
|
151 |
-
image = maintain_aspect_ratio_resize(image, height=350) # save frame as JPEG file
|
152 |
-
cv2.imwrite( img_fpath, image) # save frame as JPEG file
|
153 |
|
154 |
count += 1
|
155 |
anno.append({
|
@@ -181,6 +181,7 @@ def extract_images_and_embeds(video_id, video_path, subtitles, output, expanded=
|
|
181 |
'text': batch_list[i]['text'],
|
182 |
'image_filepath': batch_list[i]['image_filepath'],
|
183 |
'start_time': batch_list[i]['start_time'],
|
|
|
184 |
})
|
185 |
batch_list = []
|
186 |
|
@@ -195,6 +196,7 @@ def extract_images_and_embeds(video_id, video_path, subtitles, output, expanded=
|
|
195 |
'text': batch_list[i]['text'],
|
196 |
'image_filepath': batch_list[i]['image_filepath'],
|
197 |
'start_time': batch_list[i]['start_time'],
|
|
|
198 |
})
|
199 |
|
200 |
with open(os.path.join(output, 'annotations.json'), 'w') as fh:
|
@@ -203,7 +205,9 @@ def extract_images_and_embeds(video_id, video_path, subtitles, output, expanded=
|
|
203 |
with open(os.path.join(output, 'embeddings.pkl'), 'wb') as fh:
|
204 |
pickle.dump(embeddings, fh)
|
205 |
|
206 |
-
def run_query(
|
|
|
|
|
207 |
|
208 |
embeddings_filepath = os.path.join(path, 'embeddings.pkl')
|
209 |
faiss_filepath = os.path.join(path, 'faiss_index.pkl')
|
@@ -228,7 +232,14 @@ def run_query(video_id, text_query, path='/tmp'):
|
|
228 |
print('Running FAISS search')
|
229 |
_, I = faiss_index.search(emb_query, 6)
|
230 |
|
231 |
-
clip_images = [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
232 |
transcripts = [f"({embeddings[idx]['start_time']}) {embeddings[idx]['text']}" for idx in I[0]]
|
233 |
return clip_images, transcripts
|
234 |
|
@@ -253,12 +264,12 @@ def get_video_id_from_url(video_url):
|
|
253 |
return url.path.split('/')[2]
|
254 |
if url.path[:3] == '/v/':
|
255 |
return url.path.split('/')[2]
|
256 |
-
|
257 |
return None
|
258 |
|
259 |
|
260 |
def process(video_url, text_query):
|
261 |
-
tmp_dir = os.
|
262 |
video_id = get_video_id_from_url(video_url)
|
263 |
output_dir = os.path.join(tmp_dir, video_id)
|
264 |
video_file = download_video(video_url, path=output_dir)
|
@@ -303,4 +314,4 @@ with gr.Blocks() as demo:
|
|
303 |
outputs=[video_player, gallery],
|
304 |
)
|
305 |
|
306 |
-
demo.launch()
|
|
|
148 |
if success:
|
149 |
img_fname = f'{video_id}_{idx:06d}'
|
150 |
img_fpath = os.path.join(output, 'frames', img_fname + '.jpg')
|
151 |
+
# image = maintain_aspect_ratio_resize(image, height=350) # save frame as JPEG file
|
152 |
+
# cv2.imwrite( img_fpath, image) # save frame as JPEG file
|
153 |
|
154 |
count += 1
|
155 |
anno.append({
|
|
|
181 |
'text': batch_list[i]['text'],
|
182 |
'image_filepath': batch_list[i]['image_filepath'],
|
183 |
'start_time': batch_list[i]['start_time'],
|
184 |
+
'frame_no': frame_no,
|
185 |
})
|
186 |
batch_list = []
|
187 |
|
|
|
196 |
'text': batch_list[i]['text'],
|
197 |
'image_filepath': batch_list[i]['image_filepath'],
|
198 |
'start_time': batch_list[i]['start_time'],
|
199 |
+
'frame_no': frame_no,
|
200 |
})
|
201 |
|
202 |
with open(os.path.join(output, 'annotations.json'), 'w') as fh:
|
|
|
205 |
with open(os.path.join(output, 'embeddings.pkl'), 'wb') as fh:
|
206 |
pickle.dump(embeddings, fh)
|
207 |
|
208 |
+
def run_query(video_path, text_query, path='/tmp'):
|
209 |
+
|
210 |
+
vidcap = cv2.VideoCapture(video_path)
|
211 |
|
212 |
embeddings_filepath = os.path.join(path, 'embeddings.pkl')
|
213 |
faiss_filepath = os.path.join(path, 'faiss_index.pkl')
|
|
|
232 |
print('Running FAISS search')
|
233 |
_, I = faiss_index.search(emb_query, 6)
|
234 |
|
235 |
+
clip_images = []
|
236 |
+
for idx in I[0]:
|
237 |
+
frame_no = embeddings[idx]['frame_no']
|
238 |
+
vidcap.set(1, frame_no) # added this line
|
239 |
+
success, image = vidcap.read()
|
240 |
+
clip_images.append(image)
|
241 |
+
|
242 |
+
# clip_images = [embeddings[idx]['image_filepath'] for idx in I[0]]
|
243 |
transcripts = [f"({embeddings[idx]['start_time']}) {embeddings[idx]['text']}" for idx in I[0]]
|
244 |
return clip_images, transcripts
|
245 |
|
|
|
264 |
return url.path.split('/')[2]
|
265 |
if url.path[:3] == '/v/':
|
266 |
return url.path.split('/')[2]
|
267 |
+
|
268 |
return None
|
269 |
|
270 |
|
271 |
def process(video_url, text_query):
|
272 |
+
tmp_dir = os.environ.get('TMPDIR', '/tmp')
|
273 |
video_id = get_video_id_from_url(video_url)
|
274 |
output_dir = os.path.join(tmp_dir, video_id)
|
275 |
video_file = download_video(video_url, path=output_dir)
|
|
|
314 |
outputs=[video_player, gallery],
|
315 |
)
|
316 |
|
317 |
+
demo.launch(share=True)
|