Spaces:
Running
Running
Upload 5 files
Browse files- LRCN_model.h5 +3 -0
- app.py +36 -1
- prediction.py +59 -0
- requirements.txt +4 -0
- youtube_downloader.py +36 -0
LRCN_model.h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:08f4c257915c46e805cd6bd83ade5612975ba3d832f4271f9362af748e27a67f
|
3 |
+
size 946728
|
app.py
CHANGED
@@ -1,2 +1,37 @@
|
|
1 |
import streamlit as st
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
from youtube_downloader import Download
|
3 |
+
from collections import deque
|
4 |
+
from prediction import Predict
|
5 |
+
|
6 |
+
downld=Download()
|
7 |
+
|
8 |
+
st.title("Human Action Recognization [DL]")
|
9 |
+
with st.expander("Details", expanded=False):
|
10 |
+
st.write('''
|
11 |
+
The chart above shows some numbers I picked for you.
|
12 |
+
I rolled actual dice for these, so they're *guaranteed* to
|
13 |
+
be random''')
|
14 |
+
|
15 |
+
|
16 |
+
url= st.text_input("Insert Youtube Url as instructed above ")
|
17 |
+
|
18 |
+
if st.button("submit"):
|
19 |
+
title, output_dir=downld.youtube_d(url)
|
20 |
+
col1, col2 = st.columns([2, 2])
|
21 |
+
|
22 |
+
with col1:
|
23 |
+
st.write(f"Video [{title}] downloaded successfully!")
|
24 |
+
print(output_dir)
|
25 |
+
st.video(f"{output_dir}/{title}.mp4")
|
26 |
+
|
27 |
+
with col2:
|
28 |
+
st.write("wait while model is performing its task")
|
29 |
+
pred =Predict()
|
30 |
+
frames_needed=25
|
31 |
+
input_path=f"{output_dir}/{title}.mp4"
|
32 |
+
output_path=f"test_videos/{title}_output.mp4"
|
33 |
+
pred.prediction(input_path,output_path ,frames_needed)
|
34 |
+
|
35 |
+
# st.video(f"test_videos/{title}.mp4")
|
36 |
+
|
37 |
+
|
prediction.py
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import numpy as np
|
3 |
+
import math
|
4 |
+
from collections import deque
|
5 |
+
import tensorflow as tf
|
6 |
+
from tensorflow import keras
|
7 |
+
from collections import OrderedDict
|
8 |
+
|
9 |
+
|
10 |
+
from keras.models import load_model
|
11 |
+
|
12 |
+
LCRN_model = load_model('LRCN_model.h5')
|
13 |
+
|
14 |
+
class Predict:
|
15 |
+
|
16 |
+
def __init__(self) :
|
17 |
+
pass
|
18 |
+
|
19 |
+
|
20 |
+
def prediction(self,input_path, output_path,frames_needed):
|
21 |
+
video_reader=cv2.VideoCapture(input_path)
|
22 |
+
|
23 |
+
original_video_width=int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
|
24 |
+
original_video_hight=int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
25 |
+
|
26 |
+
video_writer=cv2.VideoWriter(output_path,cv2.VideoWriter_fourcc("M","P","4","V"),video_reader.get(cv2.CAP_PROP_FPS),(original_video_width,original_video_hight))
|
27 |
+
frames_deque=deque(maxlen=frames_needed)
|
28 |
+
|
29 |
+
predicted_class_name=""
|
30 |
+
IMAGE_HEIGHT , IMAGE_WIDTH = 64, 64
|
31 |
+
classes_to_be_used=["HorseRace","BenchPress","PullUps","PushUps","HorseRiding","HighJump","Swing"]
|
32 |
+
|
33 |
+
while video_reader.isOpened():
|
34 |
+
|
35 |
+
ok, frame= video_reader.read()
|
36 |
+
|
37 |
+
if not ok:
|
38 |
+
break
|
39 |
+
|
40 |
+
resized_frame=cv2.resize(frame,(IMAGE_HEIGHT,IMAGE_WIDTH))
|
41 |
+
|
42 |
+
normalized_frame= resized_frame/255
|
43 |
+
|
44 |
+
frames_deque.append(normalized_frame)
|
45 |
+
|
46 |
+
if len(frames_deque)==frames_needed:
|
47 |
+
|
48 |
+
predicted_labels_probabilities= LCRN_model.predict(np.expand_dims(frames_deque,axis=0)[0])
|
49 |
+
|
50 |
+
predicted_label=np.argmax(predicted_labels_probabilities)
|
51 |
+
|
52 |
+
predicted_class_name=classes_to_be_used[predicted_label]
|
53 |
+
|
54 |
+
cv2.putText(frame,predicted_class_name,(10,30),cv2.FONT_HERSHEY_SIMPLEX,1,(0,255,0),2)
|
55 |
+
|
56 |
+
video_writer.write(frame)
|
57 |
+
|
58 |
+
video_reader.release()
|
59 |
+
video_writer.release()
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
streamlit
|
2 |
+
!pip install pytube@git+https://github.com/priyankaj1311/pytube.git@master_copy
|
3 |
+
|
4 |
+
opencv-python
|
youtube_downloader.py
ADDED
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from pytube import YouTube
|
2 |
+
import streamlit as st
|
3 |
+
import os
|
4 |
+
|
5 |
+
# YouTube video URL
|
6 |
+
# video_url = 'https://youtu.be/7zJvNZVZMew'
|
7 |
+
|
8 |
+
# Set the output directory
|
9 |
+
|
10 |
+
|
11 |
+
# # Choose the highest resolution stream (usually the first one)
|
12 |
+
# video_stream = yt.streams.get_highest_resolution()
|
13 |
+
|
14 |
+
# # Download the video
|
15 |
+
# video_stream.download(output_path=output_dir)
|
16 |
+
|
17 |
+
# print("Video downloaded successfully!")
|
18 |
+
|
19 |
+
class Download:
|
20 |
+
def __init__(self):
|
21 |
+
pass
|
22 |
+
|
23 |
+
def youtube_d(self,url):
|
24 |
+
self.url = url
|
25 |
+
os.makedirs('test_videos',exist_ok=True)
|
26 |
+
output_dir = "test_videos"
|
27 |
+
print(url)
|
28 |
+
yt = YouTube(url)
|
29 |
+
video_stream = yt.streams.get_highest_resolution()
|
30 |
+
video_stream.download(output_path=output_dir)
|
31 |
+
|
32 |
+
return yt.title, output_dir
|
33 |
+
|
34 |
+
|
35 |
+
|
36 |
+
|