Spaces:
Runtime error
Runtime error
yiyixuxu
commited on
Commit
•
5f4ce2c
1
Parent(s):
15b3749
added batch processing for image encoding
Browse files
app.py
CHANGED
@@ -30,19 +30,24 @@ def select_video_format(url, format_note='480p', ext='mp4'):
|
|
30 |
format_id = format.get('format_id', None)
|
31 |
fps = format.get('fps', None)
|
32 |
print(f'format selected: {format}')
|
33 |
-
return(format_id, fps)
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
print(f"testing...all the files in local directory: {os.listdir('.')}")
|
38 |
ydl_opts = {
|
39 |
'format':format_id,
|
40 |
-
'outtmpl': "
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
with youtube_dl.YoutubeDL(ydl_opts) as ydl:
|
42 |
try:
|
43 |
ydl.cache.remove()
|
44 |
meta = ydl.extract_info(url)
|
45 |
-
save_location = meta['id'] + '.' + meta['ext']
|
46 |
except youtube_dl.DownloadError as error:
|
47 |
print(f'error with download_video function: {error}')
|
48 |
return(save_location)
|
@@ -51,17 +56,17 @@ def process_video_parallel(video, skip_frames, dest_path, num_processes, process
|
|
51 |
cap = cv2.VideoCapture(video)
|
52 |
frames_per_process = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) // (num_processes)
|
53 |
count = frames_per_process * process_number
|
|
|
54 |
print(f"worker: {process_number}, process frames {count} ~ {frames_per_process * (process_number + 1)} \n total number of frames: {cap.get(cv2.CAP_PROP_FRAME_COUNT)} \n video: {video}; isOpen? : {cap.isOpened()}")
|
55 |
while count < frames_per_process * (process_number + 1) :
|
56 |
ret, frame = cap.read()
|
57 |
if not ret:
|
58 |
break
|
59 |
-
count
|
60 |
-
if (count - frames_per_process * process_number) % skip_frames ==0:
|
61 |
filename =f"{dest_path}/{count}.jpg"
|
62 |
cv2.imwrite(filename, frame)
|
63 |
#print(f"saved {filename}")
|
64 |
-
|
65 |
cap.release()
|
66 |
|
67 |
|
@@ -74,9 +79,8 @@ def vid2frames(url, sampling_interval=1, ext='mp4'):
|
|
74 |
shutil.rmtree(dest_path)
|
75 |
dest_path.mkdir(parents=True)
|
76 |
# figure out the format for download,
|
77 |
-
# by default select 480p
|
78 |
-
|
79 |
-
format_id, fps = select_video_format(url, format_note='480p', ext='mp4')
|
80 |
# download the video
|
81 |
video = download_video(url,format_id)
|
82 |
# calculate skip_frames
|
@@ -85,27 +89,16 @@ def vid2frames(url, sampling_interval=1, ext='mp4'):
|
|
85 |
except:
|
86 |
skip_frames = int(30 * sampling_interval)
|
87 |
|
88 |
-
|
89 |
print(f'video saved at: {video}, fps:{fps}, skip_frames: {skip_frames}')
|
90 |
# extract video frames at given sampling interval with multiprocessing -
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
print(f'video: {video}; isOpen? : {cap.isOpened()}')
|
96 |
-
print(f'n_workers: {n_workers}')
|
97 |
with Pool(n_workers) as pool:
|
98 |
pool.map(partial(process_video_parallel, video, skip_frames, dest_path, n_workers), range(n_workers))
|
99 |
-
|
100 |
-
|
101 |
-
images = []
|
102 |
-
filenames = sorted(dest_path.glob('*.jpg'),key=lambda p: int(p.stem))
|
103 |
-
print(f"extracted {len(filenames)} frames")
|
104 |
-
for filename in filenames:
|
105 |
-
image = Image.open(filename).convert("RGB")
|
106 |
-
original_images.append(image)
|
107 |
-
images.append(preprocess(image))
|
108 |
-
return original_images, images
|
109 |
|
110 |
|
111 |
def captioned_strip(images, caption=None, times=None, rows=1):
|
@@ -116,8 +109,6 @@ def captioned_strip(images, caption=None, times=None, rows=1):
|
|
116 |
img.paste(img_, (i // rows * w, increased_h + (i % rows) * h))
|
117 |
if caption is not None:
|
118 |
draw = ImageDraw.Draw(img)
|
119 |
-
#font = ImageFont.load_default()
|
120 |
-
#font_small = ImageFont.truetype("arial.pil", 12)
|
121 |
font = ImageFont.truetype(
|
122 |
"/usr/share/fonts/truetype/liberation2/LiberationMono-Bold.ttf", 16
|
123 |
)
|
@@ -131,26 +122,40 @@ def captioned_strip(images, caption=None, times=None, rows=1):
|
|
131 |
(255, 255, 255), font=font_small)
|
132 |
return img
|
133 |
|
134 |
-
def run_inference(url, sampling_interval, search_query):
|
135 |
-
|
136 |
-
|
137 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
with torch.no_grad():
|
139 |
-
image_features = model.encode_image(image_input)
|
140 |
text_features = model.encode_text(clip.tokenize(search_query).to(device))
|
141 |
-
|
142 |
-
image_features /= image_features.norm(dim=-1, keepdim=True)
|
143 |
-
text_features /= text_features.norm(dim=-1, keepdim=True)
|
144 |
|
145 |
similarity = (100.0 * image_features @ text_features.T)
|
146 |
values, indices = similarity.topk(4, dim=0)
|
147 |
-
|
148 |
-
best_frames = [
|
149 |
times = [f'{datetime.timedelta(seconds = ind[0].item() * sampling_interval)}' for ind in indices]
|
150 |
-
print("testing... before captioned_strip func")
|
151 |
image_output = captioned_strip(best_frames,search_query, times,2)
|
152 |
title = search_query
|
153 |
-
print("testing... after captioned_strip func")
|
154 |
return(title, image_output)
|
155 |
|
156 |
inputs = [gr.inputs.Textbox(label="Give us the link to your youtube video!"),
|
|
|
30 |
format_id = format.get('format_id', None)
|
31 |
fps = format.get('fps', None)
|
32 |
print(f'format selected: {format}')
|
33 |
+
return(format, format_id, fps)
|
34 |
|
35 |
+
# to-do: delete saved videos
|
36 |
+
def download_video(url,format_id, n_keep=10):
|
|
|
37 |
ydl_opts = {
|
38 |
'format':format_id,
|
39 |
+
'outtmpl': "videos/%(id)s.%(ext)s"}
|
40 |
+
# create a directory for saved videos
|
41 |
+
video_path = Path('videos')
|
42 |
+
try:
|
43 |
+
video_path.mkdir(parents=True)
|
44 |
+
except FileExistsError:
|
45 |
+
pass
|
46 |
with youtube_dl.YoutubeDL(ydl_opts) as ydl:
|
47 |
try:
|
48 |
ydl.cache.remove()
|
49 |
meta = ydl.extract_info(url)
|
50 |
+
save_location = 'videos/' + meta['id'] + '.' + meta['ext']
|
51 |
except youtube_dl.DownloadError as error:
|
52 |
print(f'error with download_video function: {error}')
|
53 |
return(save_location)
|
|
|
56 |
cap = cv2.VideoCapture(video)
|
57 |
frames_per_process = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) // (num_processes)
|
58 |
count = frames_per_process * process_number
|
59 |
+
cap.set(cv2.CAP_PROP_POS_FRAMES, count)
|
60 |
print(f"worker: {process_number}, process frames {count} ~ {frames_per_process * (process_number + 1)} \n total number of frames: {cap.get(cv2.CAP_PROP_FRAME_COUNT)} \n video: {video}; isOpen? : {cap.isOpened()}")
|
61 |
while count < frames_per_process * (process_number + 1) :
|
62 |
ret, frame = cap.read()
|
63 |
if not ret:
|
64 |
break
|
65 |
+
if count % skip_frames ==0:
|
|
|
66 |
filename =f"{dest_path}/{count}.jpg"
|
67 |
cv2.imwrite(filename, frame)
|
68 |
#print(f"saved {filename}")
|
69 |
+
count += 1
|
70 |
cap.release()
|
71 |
|
72 |
|
|
|
79 |
shutil.rmtree(dest_path)
|
80 |
dest_path.mkdir(parents=True)
|
81 |
# figure out the format for download,
|
82 |
+
# by default select 480p and .mp4
|
83 |
+
format, format_id, fps = select_video_format(url, format_note='480p', ext='mp4')
|
|
|
84 |
# download the video
|
85 |
video = download_video(url,format_id)
|
86 |
# calculate skip_frames
|
|
|
89 |
except:
|
90 |
skip_frames = int(30 * sampling_interval)
|
91 |
|
|
|
92 |
print(f'video saved at: {video}, fps:{fps}, skip_frames: {skip_frames}')
|
93 |
# extract video frames at given sampling interval with multiprocessing -
|
94 |
+
n_workers = min(os.cpu_count(), 12)
|
95 |
+
|
96 |
+
print(f'now extracting frames with {n_workers} process...')
|
97 |
+
|
|
|
|
|
98 |
with Pool(n_workers) as pool:
|
99 |
pool.map(partial(process_video_parallel, video, skip_frames, dest_path, n_workers), range(n_workers))
|
100 |
+
return(skip_frames, dest_path)
|
101 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
102 |
|
103 |
|
104 |
def captioned_strip(images, caption=None, times=None, rows=1):
|
|
|
109 |
img.paste(img_, (i // rows * w, increased_h + (i % rows) * h))
|
110 |
if caption is not None:
|
111 |
draw = ImageDraw.Draw(img)
|
|
|
|
|
112 |
font = ImageFont.truetype(
|
113 |
"/usr/share/fonts/truetype/liberation2/LiberationMono-Bold.ttf", 16
|
114 |
)
|
|
|
122 |
(255, 255, 255), font=font_small)
|
123 |
return img
|
124 |
|
125 |
+
def run_inference(url, sampling_interval, search_query, bs=256):
|
126 |
+
skip_frames, path_frames= vid2frames(url,sampling_interval)
|
127 |
+
filenames = sorted(path_frames.glob('*.jpg'),key=lambda p: int(p.stem))
|
128 |
+
n_frames = len(filenames)
|
129 |
+
bs = min(n_frames,bs)
|
130 |
+
print(f"extracted {n_frames} frames, now encoding images")
|
131 |
+
# encoding images one batch at a time, combine all batch outputs -> image_features, size n_frames x 512
|
132 |
+
image_features = torch.empty(size=(n_frames, 512), dtype=torch.float16).to(device)
|
133 |
+
print(f"batch size :{bs} ; number of batches: {len(range(0, n_frames,bs))}")
|
134 |
+
for b in range(0, n_frames,bs):
|
135 |
+
images = []
|
136 |
+
# loop through all frames in the batch -> create batch_image_input, size bs x 3 x 224 x 224
|
137 |
+
for filename in filenames[b:b+bs]:
|
138 |
+
image = Image.open(filename).convert("RGB")
|
139 |
+
images.append(preprocess(image))
|
140 |
+
batch_image_input = torch.tensor(np.stack(images)).to(device)
|
141 |
+
# encoding batch_image_input -> batch_image_features
|
142 |
+
with torch.no_grad():
|
143 |
+
batch_image_features = model.encode_image(batch_image_input)
|
144 |
+
batch_image_features /= batch_image_features.norm(dim=-1, keepdim=True)
|
145 |
+
# add encoded image embedding to image_features
|
146 |
+
image_features[b:b+bs] = batch_image_features
|
147 |
+
# encoding search query
|
148 |
with torch.no_grad():
|
|
|
149 |
text_features = model.encode_text(clip.tokenize(search_query).to(device))
|
150 |
+
text_features /= text_features.norm(dim=-1, keepdim=True)
|
|
|
|
|
151 |
|
152 |
similarity = (100.0 * image_features @ text_features.T)
|
153 |
values, indices = similarity.topk(4, dim=0)
|
154 |
+
|
155 |
+
best_frames = [Image.open(filenames[ind]).convert("RGB") for ind in indices]
|
156 |
times = [f'{datetime.timedelta(seconds = ind[0].item() * sampling_interval)}' for ind in indices]
|
|
|
157 |
image_output = captioned_strip(best_frames,search_query, times,2)
|
158 |
title = search_query
|
|
|
159 |
return(title, image_output)
|
160 |
|
161 |
inputs = [gr.inputs.Textbox(label="Give us the link to your youtube video!"),
|