English | 简体中文

PaddleVide Quick Start

• 1. Installation
  • 1.1 Install PaddlePaddle
  • 1.2 Install PaddleVideo Whl Package
• 2. Easy-to-Use
  • 2.1 Use by Command Line
  • 2.2 Use by Python Code
• 3. Arguments description
• 4.QA

1. Installation

1.1 Install PaddlePaddle

• If you have CUDA 9 or CUDA 10 installed on your machine, please run the following command to install

  python3.7 -m pip install paddlepaddle-gpu -i https://mirror.baidu.com/pypi/simple
  

• If you have no available GPU on your machine, please run the following command to install the CPU version

  python3.7 -m pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple
  

For more software version requirements, please refer to the instructions in Installation Document for operation.

1.2 Install PaddleVideo Whl Package

• option1: use pypi（recommand）

pip3.7 install ppvideo==2.3.0

• option2: build and install locally

python3.7 setup.py bdist_wheel
python3.7 -m pip install dist/ppvideo-2.3.0-py3-none-any.whl

2. Easy-to-Use

2.1 Use by Command Line

Run shell command：

ppvideo --model_name='ppTSM_v2' --use_gpu=False --video_file='data/example.avi'

• This command use PP-TSM_v2 model to infer data/example.avi file in CPU.

• The length of the example video is about 10s. When inference, the video is first divided into 16 segments according to the time axis, then extract one frame from each segment. Finally all frames are combined and feeded into the network.

Results：

Current video file: data/example.avi
        top-1 classes: [5]
        top-1 scores: [1.]
        top-1 label names: ['archery']

As you can see, use PP-TSM_v2 trained on Kinetics-400 to predict data/example.avi video，top1 prediction class_id is 5, scores is 1.0, class name is archery.

2.2 Use by Python Code

Run python code：

from ppvideo import PaddleVideo
clas = PaddleVideo(model_name='ppTSM_v2', use_gpu=False)
video_file='data/example.avi'
clas.predict(video_file)

• This code use PP-TSM_v2 model to infer data/example.avi file in CPU.

Results:

Current video file: data/example.avi
        top-1 classes: [5]
        top-1 scores: [1.]
        top-1 label names: ['archery']

As you can see, use PP-TSM_v2 trained on Kinetics-400 to predict data/example.avi video，top1 prediction class_id is 5, scores is 1.0, class name is archery.

3. Arguments description

command example1：

ppvideo --model_name='ppTSM_v2' --num_seg=16 --video_file="data/mp4" --batch_size=2  --top_k=5

Results：

Current video file: data/mp4/example3.avi
        top-5 classes: [  5 345 311 159 327]
        top-5 scores: [1.0000000e+00 1.0152016e-11 8.2871061e-14 6.7713670e-14 5.0752070e-14]
        top-5 label names: ['archery', 'sword_fighting', 'skipping_rope', 'hula_hooping', 'spray_painting']
Current video file: data/mp4/example2.avi
        top-5 classes: [  5 345 311 159 327]
        top-5 scores: [1.0000000e+00 1.0152016e-11 8.2871061e-14 6.7713670e-14 5.0752070e-14]
        top-5 label names: ['archery', 'sword_fighting', 'skipping_rope', 'hula_hooping', 'spray_painting']
Current video file: data/mp4/example.avi
        top-5 classes: [  5 345 311 159 327]
        top-5 scores: [1.0000000e+00 1.0152016e-11 8.2871061e-14 6.7713670e-14 5.0752070e-14]
        top-5 label names: ['archery', 'sword_fighting', 'skipping_rope', 'hula_hooping', 'spray_painting']
Current video file: data/mp4/example1.avi
        top-5 classes: [  5 345 311 159 327]
        top-5 scores: [1.0000000e+00 1.0152016e-11 8.2871061e-14 6.7713670e-14 5.0752070e-14]
        top-5 label names: ['archery', 'sword_fighting', 'skipping_rope', 'hula_hooping', 'spray_painting']

command example1：

ppvideo --model_name='ppTSM' --num_seg=8 --video_file="data/mp4" --batch_size=2  --top_k=5

4. QA

1. opecv-python Installation maybe slow, you can try:

python3.7 -m pip install opencv-python==4.2.0.32 -i https://pypi.doubanio.com/simple