Update app.py
Browse files
app.py
CHANGED
@@ -21,39 +21,11 @@ import numpy as np
|
|
21 |
import torch
|
22 |
import matplotlib.pyplot as plt
|
23 |
import torchvision.transforms.functional as F
|
|
|
|
|
|
|
24 |
|
25 |
|
26 |
-
plt.rcParams["savefig.bbox"] = "tight"
|
27 |
-
# sphinx_gallery_thumbnail_number = 2
|
28 |
-
|
29 |
-
|
30 |
-
def plot(imgs, **imshow_kwargs):
|
31 |
-
if not isinstance(imgs[0], list):
|
32 |
-
# Make a 2d grid even if there's just 1 row
|
33 |
-
imgs = [imgs]
|
34 |
-
|
35 |
-
num_rows = len(imgs)
|
36 |
-
num_cols = len(imgs[0])
|
37 |
-
_, axs = plt.subplots(nrows=num_rows, ncols=num_cols, squeeze=False)
|
38 |
-
for row_idx, row in enumerate(imgs):
|
39 |
-
for col_idx, img in enumerate(row):
|
40 |
-
ax = axs[row_idx, col_idx]
|
41 |
-
img = F.to_pil_image(img.to("cpu"))
|
42 |
-
ax.imshow(np.asarray(img), **imshow_kwargs)
|
43 |
-
ax.set(xticklabels=[], yticklabels=[], xticks=[], yticks=[])
|
44 |
-
|
45 |
-
plt.tight_layout()
|
46 |
-
|
47 |
-
###################################
|
48 |
-
# Reading Videos Using Torchvision
|
49 |
-
# --------------------------------
|
50 |
-
# We will first read a video using :func:`~torchvision.io.read_video`.
|
51 |
-
# Alternatively one can use the new :class:`~torchvision.io.VideoReader` API (if
|
52 |
-
# torchvision is built from source).
|
53 |
-
# The video we will use here is free of use from `pexels.com
|
54 |
-
# <https://www.pexels.com/video/a-man-playing-a-game-of-basketball-5192157/>`_,
|
55 |
-
# credits go to `Pavel Danilyuk <https://www.pexels.com/@pavel-danilyuk>`_.
|
56 |
-
|
57 |
|
58 |
import tempfile
|
59 |
from pathlib import Path
|
@@ -64,29 +36,15 @@ def infer():
|
|
64 |
video_path = Path(tempfile.mkdtemp()) / "basketball.mp4"
|
65 |
_ = urlretrieve(video_url, video_path)
|
66 |
|
67 |
-
#########################
|
68 |
-
# :func:`~torchvision.io.read_video` returns the video frames, audio frames and
|
69 |
-
# the metadata associated with the video. In our case, we only need the video
|
70 |
-
# frames.
|
71 |
-
#
|
72 |
-
# Here we will just make 2 predictions between 2 pre-selected pairs of frames,
|
73 |
-
# namely frames (100, 101) and (150, 151). Each of these pairs corresponds to a
|
74 |
-
# single model input.
|
75 |
|
76 |
-
from torchvision.io import read_video
|
77 |
-
frames, _, _ = read_video(str(video_path), output_format="TCHW")
|
78 |
|
79 |
-
|
80 |
-
|
81 |
|
82 |
-
|
83 |
-
|
84 |
-
# :func:`~torchvision.io.read_video` and resize them to ensure their
|
85 |
-
# dimensions are divisible by 8. Then we use the transforms bundled into the
|
86 |
-
# weights in order to preprocess the input and rescale its values to the
|
87 |
-
# required ``[-1, 1]`` interval.
|
88 |
|
89 |
-
|
90 |
|
91 |
weights = Raft_Large_Weights.DEFAULT
|
92 |
transforms = weights.transforms()
|
@@ -112,7 +70,7 @@ def infer():
|
|
112 |
# We also provide the :func:`~torchvision.models.optical_flow.raft_small` model
|
113 |
# builder, which is smaller and faster to run, sacrificing a bit of accuracy.
|
114 |
|
115 |
-
|
116 |
|
117 |
# If you can, run this example on a GPU, it will be a lot faster.
|
118 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
21 |
import torch
|
22 |
import matplotlib.pyplot as plt
|
23 |
import torchvision.transforms.functional as F
|
24 |
+
from torchvision.io import read_video
|
25 |
+
from torchvision.models.optical_flow import Raft_Large_Weights
|
26 |
+
from torchvision.models.optical_flow import raft_large
|
27 |
|
28 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
import tempfile
|
31 |
from pathlib import Path
|
|
|
36 |
video_path = Path(tempfile.mkdtemp()) / "basketball.mp4"
|
37 |
_ = urlretrieve(video_url, video_path)
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
|
|
|
|
40 |
|
41 |
+
|
42 |
+
frames, _, _ = read_video(str(video_path), output_format="TCHW")
|
43 |
|
44 |
+
img1= frames[100]
|
45 |
+
img2 = frames[101]
|
|
|
|
|
|
|
|
|
46 |
|
47 |
+
|
48 |
|
49 |
weights = Raft_Large_Weights.DEFAULT
|
50 |
transforms = weights.transforms()
|
|
|
70 |
# We also provide the :func:`~torchvision.models.optical_flow.raft_small` model
|
71 |
# builder, which is smaller and faster to run, sacrificing a bit of accuracy.
|
72 |
|
73 |
+
|
74 |
|
75 |
# If you can, run this example on a GPU, it will be a lot faster.
|
76 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|