!git clone https://github.com/facebookresearch/dino.git /home/featurize/dino

Cloning into '/home/featurize/dino'...
remote: Enumerating objects: 168, done.
remote: Counting objects: 100% (24/24), done.
remote: Compressing objects: 100% (19/19), done.
remote: Total 168 (delta 10), reused 16 (delta 5), pack-reused 144
Receiving objects: 100% (168/168), 24.47 MiB | 10.41 MiB/s, done.
Resolving deltas: 100% (98/98), done.

## 目录准备以及下载图片样例

!mkdir /home/featurize/dino/input
!mkdir /home/featurize/dino/output
!wget https://www.snowskool.com/uploads/images/ski_and_snowboard_snowsport12.jpg -O /home/featurize/dino/input/ski.jpg

import matplotlib.pyplot as plt
import cv2

plt.axis('off')
plt.imshow(cv2.cvtColor(cv2.imread('/home/featurize/dino/input/ski.jpg'), cv2.COLOR_BGR2RGB));

mkdir: cannot create directory ‘/home/featurize/dino/input’: File exists
mkdir: cannot create directory ‘/home/featurize/dino/output’: File exists
--2021-11-30 14:38:32--  https://www.snowskool.com/uploads/images/ski_and_snowboard_snowsport12.jpg
Connecting to 172.16.0.13:7890... connected.
Proxy request sent, awaiting response... 200 OK
Length: 110231 (108K) [image/jpeg]
Saving to: ‘/home/featurize/dino/input/ski.jpg’

/home/featurize/din 100%[===================>] 107.65K   347KB/s    in 0.3s    

2021-11-30 14:38:34 (347 KB/s) - ‘/home/featurize/dino/input/ski.jpg’ saved [110231/110231]

!python /home/featurize/dino/visualize_attention.py  \
--image_path /home/featurize/dino/input/ski.jpg  \
--output_dir /home/featurize/dino/output

Please use the `--pretrained_weights` argument to indicate the path of the checkpoint to evaluate.
Since no pretrained weights have been provided, we load the reference pretrained DINO weights.
Downloading: "https://dl.fbaipublicfiles.com/dino/dino_deitsmall8_300ep_pretrain/dino_deitsmall8_300ep_pretrain.pth" to /home/featurize/.cache/torch/hub/checkpoints/dino_deitsmall8_300ep_pretrain.pth
100%|██████████████████████████████████████| 82.7M/82.7M [00:08<00:00, 10.4MB/s]
/environment/python/versions/3.7.12/lib/python3.7/site-packages/torch/nn/functional.py:3635: UserWarning: Default upsampling behavior when mode=bicubic is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details.
  "See the documentation of nn.Upsample for details.".format(mode)
/environment/python/versions/3.7.12/lib/python3.7/site-packages/torch/nn/functional.py:3680: UserWarning: The default behavior for interpolate/upsample with float scale_factor changed in 1.6.0 to align with other frameworks/libraries, and now uses scale_factor directly, instead of relying on the computed output size. If you wish to restore the old behavior, please set recompute_scale_factor=True. See the documentation of nn.Upsample for details. 
  "The default behavior for interpolate/upsample with float scale_factor changed "
/home/featurize/dino/output/attn-head0.png saved.
/home/featurize/dino/output/attn-head1.png saved.
/home/featurize/dino/output/attn-head2.png saved.
/home/featurize/dino/output/attn-head3.png saved.
/home/featurize/dino/output/attn-head4.png saved.
/home/featurize/dino/output/attn-head5.png saved.

f, axs = plt.subplots(2,3, figsize=(12,8))
for i, ax in enumerate(axs.reshape(-1)):
    ax.axis('off')
    ax.imshow(cv2.cvtColor(cv2.imread(f'/home/featurize/dino/output/attn-head{i}.png'), cv2.COLOR_BGR2RGB));

# 下载测试视频
!unset http_proxy;unset https_proxy;unset all_proxy
!wget https://featurize.oss-cn-chengdu.aliyuncs.com/input.mp4 -O /home/featurize/dino/input/input.mp4

# Jupyter里可视化一下
from IPython.display import Video
Video('/home/featurize/dino/input/input.mp4', embed=True, width=800, height=600)

# DINO 进行推理
!python /home/featurize/dino/video_generation.py  \
--pretrained_weights dino_deitsmall8_pretrain.pth  \
--input_path /home/featurize/dino/input/input.mp4  \
--output_path /home/featurize/dino/output \
--video_format mp4 \
--fps 25

Please use the `--pretrained_weights` argument to indicate the path of the checkpoint to evaluate.
Since no pretrained weights have been provided, we load the reference pretrained DINO weights.
Video: /home/featurize/dino/input/input.mp4 (24.978432290124694 fps)
Extracting frames to /home/featurize/dino/output/frames
Generating attention images to /home/featurize/dino/output/attention
  0%|                                                   | 0/613 [00:00<?, ?it/s]/environment/python/versions/3.7.12/lib/python3.7/site-packages/torch/nn/functional.py:3635: UserWarning: Default upsampling behavior when mode=bicubic is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details.
  "See the documentation of nn.Upsample for details.".format(mode)
/environment/python/versions/3.7.12/lib/python3.7/site-packages/torch/nn/functional.py:3680: UserWarning: The default behavior for interpolate/upsample with float scale_factor changed in 1.6.0 to align with other frameworks/libraries, and now uses scale_factor directly, instead of relying on the computed output size. If you wish to restore the old behavior, please set recompute_scale_factor=True. See the documentation of nn.Upsample for details. 
  "The default behavior for interpolate/upsample with float scale_factor changed "
100%|█████████████████████████████████████████| 613/613 [11:57<00:00,  1.17s/it]
Generating video (1280, 720) to /home/featurize/dino/output
100%|█████████████████████████████████████████| 612/612 [00:07<00:00, 85.13it/s]
OpenCV: FFMPEG: tag 0x5634504d/'MP4V' is not supported with codec id 12 and format 'mp4 / MP4 (MPEG-4 Part 14)'
OpenCV: FFMPEG: fallback to use tag 0x7634706d/'mp4v'
Done

# 这是 Dave 之前推理的视频
Video('https://featurize.oss-cn-chengdu.aliyuncs.com/output.mp4')