update to latest with Daan's changes

app.py

@@ -90,7 +90,8 @@ transfer_data_indices_to_temp() # NOTE: Only works after doing 'git lfs pull' to
 with open('apb2022.json', "r") as json_file:
     TARGET_ENTRIES = json.load(json_file)
 
-EXAMPLE_VIDEO_URLS = ["https://drive.google.com/uc?id=1Y1-ypXOvLrp1x0cjAe_hMobCEdA0UbEo&export=download",
+EXAMPLE_VIDEO_URLS = ["https://www.youtube.com/watch?v=qIaqMqMweM4",
+                      "https://drive.google.com/uc?id=1Y1-ypXOvLrp1x0cjAe_hMobCEdA0UbEo&export=download",
                       "https://video.twimg.com/amplify_video/1575576025651617796/vid/480x852/jP057nPfPJSUM0kR.mp4?tag=14",
                       "https://drive.google.com/uc?id=1XW0niHR1k09vPNv1cp6NvdGXe7FHJc1D&export=download"]
 

config.py

@@ -5,5 +5,5 @@ VIDEO_DIRECTORY = tempfile.gettempdir()
 
 FPS = 5
 MIN_DISTANCE = 4
-MAX_DISTANCE = 30 # Used to be 30
+MAX_DISTANCE = 50 # Used to be 30
 ROLLING_WINDOW_SIZE = 10

data/0fd24e056447f6cd068f40ba2e41828c.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d304bf65b174cc6d446536e183062988d13450eb27d6e1490b6024813cfca61
+oid sha256:beba3b8f175ccd3b8fc74c32524d013f85dc0c19fcd467d8e13d75381886f4a9
 size 18035

data/1de5e559d3d190ca09ac3fa6b40a0138.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a81da7184d4a73b80512031805e28d76adbf45c508b16cde173a9cc980104b48
+oid sha256:31a5c8e5d2b8c433f2a570059ee680c1fddd22118915cbe3e964f9019730d238
 size 5395

data/265732a300bbe18c4d39df108fe0c121.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3d30c03a7817befd97b670f124e80bfcd2c6862a3bf3d0a57bdb7c5cad7122b
+oid sha256:9ab0ada81fdd3e5ab2e7e3d276c38dabf052c2c0d289dd3021da0406b892bbbb
 size 284155

data/2909574cb590231c5d1028032bd849a8.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8634dffda2a11307f55712a079bba4e0e38895f9f3836d26c5a32a051580cfcd
+oid sha256:060b6cfb3c748275d824866bed83033a9b75234d215c36d9a04454d41c7d6697
 size 6287235

data/2f71b6c1d5c91c4b73eef2ebd05bd848.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4357ef337a87f67f43283e770eb444c2261f3233c64d6aef04f671a1b4237780
+oid sha256:5f8df69be9a6eb7c614791f6d06d0c341d7f063d74031e064fafa99c08df6c0c
 size 3725195

data/75617545ba72efdc9a7b2b140184a9c6.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c28e09d9cee38776ad698e56209573f2e3f9f67a533b10b9001a25b39e34321e
+oid sha256:03d76445b9fad00277f34890930f3aae6d6e078b3308ea18c83ad065db0d1754
 size 3305875

data/9c0a24aa59358ee014ded54c88eaa6c1.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35786ba354b46631f880ff468b04f234dd0a8fff1ea159385d9a252272a53595
+oid sha256:44e714afd9b69d1104adb7192352d6a3b8070510f03840fbb2fc9f2b95ccffac
 size 294515

data/a3efa8f2e01e0964d7987e6b1e4b3b44.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ab33d752701be55ece43a841e0318a7a8eb717c9fb3c09c9e9cccd57528f65b
+oid sha256:54185972343a709462b8e72ad79d0e7cb3bf33aff65c92f5d65ca8b8c0e73c39
 size 5395

data/b3cf94671de7aadc22970483150aafd5.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30157e51167c48c9e9da8400bd2efb4db730cf99060ec6b958e593caf4b934ca
+oid sha256:82e10462195550af464d3e4c2488712dd01b3317dd275c9caf6cea94ad80e4a3
 size 3075

data/d0369414718aa9573ff9d009f4f61cc0.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da68a19cadf2abcba5da6660393547599ae3a70f73c683ce236962f44a6f24ce
+oid sha256:96c567a4fb8b625441781820ebdc7db7835fc0a73dd6c54c0d7ad540428038d3
 size 6520355

requirements.txt

@@ -6,4 +6,6 @@ faiss-cpu==1.7.2
 Pillow==9.2.0
 kats==0.2.0
 seaborn==0.12.0
-scipy==1.7.3
+scipy==1.7.3
+pytube==12.1.0
+pytube3==9.6.4

videohash.py

@@ -3,11 +3,14 @@ import urllib.request
 import shutil
 import logging
 import hashlib
+import time
 
 from PIL import Image
 import imagehash
 from moviepy.editor import VideoFileClip
-import numpy as np  
+from moviepy.video.fx.all import crop
+import numpy as np
+from pytube import YouTube
 
 from config import FPS, VIDEO_DIRECTORY
 
@@ -18,12 +21,26 @@ def filepath_from_url(url):
 
 def download_video_from_url(url):
     """Download video from url or return md5 hash as video name"""
-    # TODO: Make work for Google link
+    start = time.time()
     filepath = filepath_from_url(url)
+
+    # Check if it exists already
     if not os.path.exists(filepath):
+        # For YouTube links
+        if url.startswith('https://www.youtube.com') or url.startswith('youtube.com') or url.startswith('http://www.youtube.com'):
+            file_dir = '/'.join(x for x in filepath.split('/')[:-1])
+            filename = filepath.split('/')[-1]
+            logging.info(f"file_dir = {file_dir}")
+            logging.info(f"filename = {filename}")
+            YouTube(url).streams.get_highest_resolution().download(file_dir, skip_existing = False, filename = filename)
+            logging.info(f"Downloaded YouTube video from {url} to {filepath} in {time.time() - start:.1f} seconds.")
+            return filepath
+
+        # Works for basically all links, except youtube 
         with (urllib.request.urlopen(url)) as f, open(filepath, 'wb') as fileout:
-            shutil.copyfileobj(f, fileout, length=16*1024)
-        logging.info(f"Downloaded video from {url} to {filepath}.")
+            logging.info(f"Starting copyfileobj on {f}")
+            shutil.copyfileobj(f, fileout, length=16*1024*1024)
+        logging.info(f"Downloaded video from {url} to {filepath} in {time.time() - start:.1f} seconds.")
     else:
         logging.info(f"Skipping downloading from {url} because {filepath} already exists.")
     return filepath
@@ -41,6 +58,13 @@ def change_ffmpeg_fps(clip, fps=FPS):
     clip.reader.lastread = clip.reader.read_frame()
     return clip
 
+def crop_video(clip, crop_percentage=0.75, w=224, h=224):
+    # Original width and height- which combined with crop_percentage determines the size of the new video
+    ow, oh = clip.size 
+
+    logging.info(f"Cropping and resizing video to ({w}, {h})")
+    return crop(clip, x_center=ow/2, y_center=oh/2, width=int(ow*crop_percentage), height=int(crop_percentage*oh)).resize((w,h))
+
 def compute_hash(frame, hash_size=16):
     image = Image.fromarray(np.array(frame))
     return imagehash.phash(image, hash_size)
@@ -51,10 +75,11 @@ def binary_array_to_uint8s(arr):
 
 def compute_hashes(url: str, fps=FPS):
     try:
-        clip = VideoFileClip(download_video_from_url(url))
+        filepath = download_video_from_url(url)
+        clip = crop_video(VideoFileClip(filepath))
     except IOError:
         logging.warn(f"Falling back to direct streaming from {url} because the downloaded video failed.")
-        clip = VideoFileClip(url)
+        clip = crop_video(VideoFileClip(url))
         
     for index, frame in enumerate(change_ffmpeg_fps(clip, fps).iter_frames()):
         # Each frame is a triplet of size (height, width, 3) of the video since it is RGB