# video-recognition-pytorch

**Repository Path**: li554/video-recognition-pytorch

## Basic Information

- **Project Name**: video-recognition-pytorch
- **Description**: 基于Pytorch实现的双流视频动作识别
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-08-06
- **Last Updated**: 2024-10-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# video-recognition-pytorch

## Introduction

This repo implements the TSM and Two Stream with shared module, attention fusion and gated weight fusion. We mainly
train the two model which use vgg16 and resnet50 as backbone on UCF101 and HMDB51 datasets.
**More models and datasets will be available later!**

## Installation

The code was tested with Anaconda and Python 3.8. After installing the Anaconda environment:

0. Clone the repo:
    ```Shell
    git clone https://gitee.com/li554/video-recognition-pytorch.git
    cd pytorch-video-recognition
    ```

1. Install dependencies:

   For PyTorch dependency, see [pytorch.org](https://pytorch.org/) for more details.

   For custom dependencies:
    ```Shell
    pip install opencv-python
    pip install tqdm scikit-learn tensorboardX
    ```

2. Download pretrained model from [BaiduYun](https://pan.baidu.com/s/1saNqGBkzZHwZpG-A5RDLVw) or
   [GoogleDrive](https://drive.google.com/file/d/19NWziHWh1LgCcHU34geoKwYezAogv9fX/view?usp=sharing). Currently only
   support pretrained model for TSMS and TSS based on resnet50.


3. You can choose different models and datasets in
   [train.py](https://github.com/li554/pytorch-video-recognition/blob/master/train.py).

   To train the model, please do:
    ```Shell
    python main.py
    ```
   specific params setting please
   check [train.py](https://github.com/li554/pytorch-video-recognition/blob/master/train.py).

## Datasets:

We used two different datasets: UCF101 and HMDB. we provide our preprocessed UCF101 and HMDB51 dataset, you can get them
here. After download done, you can unzip the dataset
in [datasets](https://github.com/li554/pytorch-video-recognition/blob/master/datasets)
or put it in other places you want while requiring you change the data command param in train.py to be your path

## Experiments

These models were trained in machine with single NVIDIA RTX 3060.