【slowfast 减少ava数据集】将ava数据集缩小到2个，对数据集做训练，然后进行检测，为训练自己的数据集做准备

谁践踏了优雅 2022-09-03 05:25 449阅读 0赞

### 目录 ###

*  [前言][Link 1]
 *  [一，数据文件结构][Link 2]
 *   *  [1.1 myava][]
     *  [1.2 annotations][]
     *  [1.3 annotations文件处理][1.3 annotations]
     *  [1.4 frame\_lists][1.4 frame_lists]
     *  [1.5 frames][]
 *  [二，预训练模型][Link 3]
 *  [四，配置文件][Link 4]
 *   *  [4.1 创建新的yaml文件][4.1 _yaml]
     *  [4.2 yaml文件解释][4.2 yaml]
 *  [五，训练][Link 5]
 *  [六，结果查看][Link 6]

# 前言 #

[b站讲解][b]  
ava的数据集非常大，训练起来非常慢，这次我就把ava的数据集缩小到2个视频（原本有299个视频），这样做的目的是观察slowfast是如何训练ava数据集的。也为我后面制作自己的数据集做下铺垫。

# 一，数据文件结构 #

首先需要把原有的数据文件的内容进行修改，原来的数据文件针对的是299个视频图片，这次只针对2个视频图片。

## 1.1 myava ##

新建一个文件夹`myava`，这里存放下图显示的文件夹：`annotations`、`frame_lists`、`frames`。

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70]

## 1.2 annotations ##

在annotations文件下，有如下文件，这些文件是从[官网][Link 7]下载的

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 1]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 2]

这里不要使用官网的默认版本，采用最新的2.2版本，官网有一个下载的链接：[https://dl.fbaipublicfiles.com/pyslowfast/annotation/ava/ava\_annotations.tar][https_dl.fbaipublicfiles.com_pyslowfast_annotation_ava_ava_annotations.tar]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 3]

下载后，解压（我是在ubuntu下解压的），使用tree查看其结构：

├── ava_annotations
    │   ├── ava_action_list_v2.1_for_activitynet_2018.pbtxt
    │   ├── ava_action_list_v2.2_for_activitynet_2019.pbtxt
    │   ├── ava_action_list_v2.2.pbtxt
    │   ├── ava_included_timestamps_v2.2.txt
    │   ├── ava_test_excluded_timestamps_v2.1.csv
    │   ├── ava_test_excluded_timestamps_v2.2.csv
    │   ├── ava_test_v2.2.csv
    │   ├── ava_train_excluded_timestamps_v2.1.csv
    │   ├── ava_train_excluded_timestamps_v2.2.csv
    │   ├── ava_train_v2.1.csv
    │   ├── ava_train_v2.2.csv
    │   ├── ava_val_excluded_timestamps_v2.1.csv
    │   ├── ava_val_excluded_timestamps_v2.2.csv
    │   ├── ava_val_v2.1.csv
    │   ├── ava_val_v2.2.csv
    │   ├── person_box_67091280_iou75
    │   │   ├── ava_detection_test_boxes_and_labels.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative_v2.1.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative_v2.2.csv
    │   │   ├── ava_detection_val_boxes_and_labels.csv
    │   │   ├── ava_detection_val_for_training_boxes_and_labels_include_negative.csv
    │   │   └── ava_detection_val_for_training_boxes_and_labels_include_negative_v2.2.csv
    │   ├── person_box_67091280_iou90
    │   │   ├── ava_action_list_v2.1_for_activitynet_2018.pbtxt
    │   │   ├── ava_detection_test_boxes_and_labels.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative_v2.1.csv
    │   │   ├── ava_detection_train_boxes_and_labels_include_negative_v2.2.csv
    │   │   ├── ava_detection_val_boxes_and_labels.csv
    │   │   ├── ava_detection_val_for_training_boxes_and_labels_include_negative.csv
    │   │   ├── ava_detection_val_for_training_boxes_and_labels_include_negative_v2.1.csv
    │   │   ├── ava_detection_val_for_training_boxes_and_labels_include_negative_v2.2.csv
    │   │   ├── ava_train_predicted_boxes.csv
    │   │   ├── ava_train_v2.1.csv
    │   │   ├── ava_val_excluded_timestamps_v2.1.csv
    │   │   ├── ava_val_predicted_boxes.csv -> ava_detection_val_boxes_and_labels.csv
    │   │   ├── ava_val_v2.1.csv
    │   │   ├── test.csv
    │   │   ├── train.csv
    │   │   └── val.csv
    │   ├── test.csv
    │   ├── train.csv
    │   └── val.csv
    └── ava_annotations.tar

这些文件不能全要，只去我们所需的，因为我们用的每一个都是需要进行修改的。

下面是需要的文件：  
`annotations`  
—`person_box_67091280_iou90`  
\------`ava_detection_train_boxes_and_labels_include_negative_v2.2.csv`  
\------`ava_detection_val_boxes_and_labels.csv`  
—`ava_action_list_v2.2_for_activitynet_2019.pbtxt`  
—`ava_detection_val_boxes_and_labels.csv`  
—`ava_train_v2.1.csv`  
—`ava_train_v2.2.csv`  
—`ava_val_excluded_timestamps_v2.2.csv`  
—`ava_val_v2.2.csv`

## 1.3 annotations文件处理 ##

每一个文件都需要处理，下面开始处理：  
1,`/annotations/person_box_67091280_iou90/ava_detection_train_boxes_and_labels_include_negative_v2.2.csv`:  
如下图所示，只要第一列名为： -5KQ66BBWC4的所有行  
![ava\_detection\_train\_boxes\_and\_labels\_include\_negative\_v2.2.cs][ava_detection_train_boxes_and_labels_include_negative_v2.2.cs]

2,`/annotations/person_box_67091280_iou90/ava_detection_val_boxes_and_labels.csv`  
如下图所示，只要第一列为：1j20qq1JyX4 的所有行  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 4]  
3,`/annotations/ava_action_list_v2.2_for_activitynet_2019.pbtxt`:  
这个文件不改，这是用来记录行为种类的，共有80个

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 5]

4,`annotations/ava_detection_val_boxes_and_labels.csv`

如下图所示，只要第一列为：1j20qq1JyX4 的所有行  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 6]

5,`/annotations/ava_train_v2.1.csv`, `/annotations/ava_train_v2.2csv`  
这两个文件非常类似，处理方法也一样

如下图所示，只要第一列名为： -5KQ66BBWC4的所有行  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 7]

6,`/annotations/ava_val_excluded_timestamps_v2.2.csv`  
这个是空文件，因为在我们所选的视频中，没有需要排除的帧。

7，`/annotations/ava_val_v2.2.csv`

如下图所示，只要第一列为：1j20qq1JyX4 的所有行

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 8]

## 1.4 frame\_lists ##

在`frame_lists`目录下，这里只存放了两个文件：`train.csv`、`val.csv`  
下载地址：[train.csv][]、[val.csv][]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 9]

同样需要修改，同上一小节讲的。

1,`frame_lists/train.csv`

2,`frame_lists/val.csv`

## 1.5 frames ##

`/frames`下有两个文件夹：`-5KQ66BBWC4`,`1j20qq1JyX4`  
这两个文件夹里存放的就是ava剪辑后的图片，制作过程参考[官网][Link 7]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 12]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 13]

# 二，预训练模型 #

最好使用预训练模型，这样可以缩短训练的一个时间，我用的预训练模型如下图

[模型下载官网][Link 8]，[预训练模型下载链接][Link 9]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 14]  
下载下来后，放在文件夹`/SlowFast/configs/AVA/c2/`下面，如下图（由于下载的预训练模型和之前一个模型重名，所以我在预训练模型后面加了个ss）  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 15]

# 四，配置文件 #

## 4.1 创建新的yaml文件 ##

在`/SlowFast/configs/AVA/`下创建一个新的yaml文件：`SLOWFAST_32x2_R50_SHORT4.yaml`，如下图

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 16]

代码如下：

TRAIN:
      ENABLE: True
      DATASET: ava
      BATCH_SIZE: 2 #64
      EVAL_PERIOD: 5
      CHECKPOINT_PERIOD: 1
      AUTO_RESUME: True
      CHECKPOINT_FILE_PATH: '/home/lxn/0yangfan/Slowfast2/SlowFast-master/configs/AVA/c2/SLOWFAST_32x2_R101_50_50ss.pkl'  #path to pretrain model
      CHECKPOINT_TYPE: caffe2
    DATA:
      NUM_FRAMES: 32
      SAMPLING_RATE: 2
      TRAIN_JITTER_SCALES: [256, 320]
      TRAIN_CROP_SIZE: 224
      TEST_CROP_SIZE: 224
      INPUT_CHANNEL_NUM: [3, 3]
      PATH_TO_DATA_DIR: '/disk6T/ava'
    DETECTION:
      ENABLE: True
      ALIGNED: True
    AVA:
      FRAME_DIR: '/disk6T/myava/frames'
      FRAME_LIST_DIR: '/disk6T/myava/frame_lists'
      ANNOTATION_DIR: '/disk6T/myava/annotations'
      #LABEL_MAP_FILE: 'ava_action_list_v2.1_for_activitynet_2018.pbtxt'
      #0GROUNDTRUTH_FILE: 'ava_val_v2.1.csv'
      #TRAIN_GT_BOX_LISTS: ['ava_train_v2.1.csv']
      DETECTION_SCORE_THRESH: 0.8
      TRAIN_PREDICT_BOX_LISTS: [
        "ava_train_v2.2.csv",
        "person_box_67091280_iou90/ava_detection_train_boxes_and_labels_include_negative_v2.2.csv",
      ]
      #TRAIN_PREDICT_BOX_LISTS: ["ava_train_predicted_boxes.csv"]
      TEST_PREDICT_BOX_LISTS: ["person_box_67091280_iou90/ava_detection_val_boxes_and_labels.csv"]
      #TEST_PREDICT_BOX_LISTS: ["ava_test_predicted_boxes.csv"]
      #EXCLUSION_FILE: "ava_train_excluded_timestamps_v2.1.csv"
      
      
    SLOWFAST:
      ALPHA: 4
      BETA_INV: 8
      FUSION_CONV_CHANNEL_RATIO: 2
      FUSION_KERNEL_SZ: 7
    RESNET:
      ZERO_INIT_FINAL_BN: True
      WIDTH_PER_GROUP: 64
      NUM_GROUPS: 1
      DEPTH: 50
      TRANS_FUNC: bottleneck_transform
      STRIDE_1X1: False
      NUM_BLOCK_TEMP_KERNEL: [[3, 3], [4, 4], [6, 6], [3, 3]]
      SPATIAL_DILATIONS: [[1, 1], [1, 1], [1, 1], [2, 2]]
      SPATIAL_STRIDES: [[1, 1], [2, 2], [2, 2], [1, 1]]
    NONLOCAL:
      LOCATION: [[[], []], [[], []], [[], []], [[], []]]
      GROUP: [[1, 1], [1, 1], [1, 1], [1, 1]]
      INSTANTIATION: dot_product
      POOL: [[[1, 2, 2], [1, 2, 2]], [[1, 2, 2], [1, 2, 2]], [[1, 2, 2], [1, 2, 2]], [[1, 2, 2], [1, 2, 2]]]
    BN:
      USE_PRECISE_STATS: False
      NUM_BATCHES_PRECISE: 200
    SOLVER:
      BASE_LR: 0.1
      LR_POLICY: steps_with_relative_lrs
      STEPS: [0, 10, 15, 20]
      LRS: [1, 0.1, 0.01, 0.001]
      MAX_EPOCH: 20
      MOMENTUM: 0.9
      WEIGHT_DECAY: 1e-7
      WARMUP_EPOCHS: 5.0
      WARMUP_START_LR: 0.000125
      OPTIMIZING_METHOD: sgd
    MODEL:
      NUM_CLASSES: 80
      ARCH: slowfast
      MODEL_NAME: SlowFast
      LOSS_FUNC: bce
      DROPOUT_RATE: 0.5
      HEAD_ACT: sigmoid
    TEST:
      ENABLE: False
      DATASET: ava
      BATCH_SIZE: 8
    DATA_LOADER:
      NUM_WORKERS: 2
      PIN_MEMORY: True
    NUM_GPUS: 1
    NUM_SHARDS: 1
    RNG_SEED: 0
    OUTPUT_DIR: .

## 4.2 yaml文件解释 ##

1.  TRAIN  
    1.1. ENABLE: True。这里将TRAIN设置为TRUE，同样的，也要TEST.ENABLE设置为False（我们只需要训练的过程）  
    1.2 BATCH\_SIZE: 2 \#64.这里batch\_size是由于我的电脑显存不够，只能设置为2，如果大家的显存够大，可以把这个batch\_size设置的大一些。  
    1.3 CHECKPOINT\_FILE\_PATH: ‘/home/lxn/0yangfan/Slowfast2/SlowFast-master/configs/AVA/c2/SLOWFAST\_32x2\_R101\_50\_50ss.pkl’ 这里放的是预训练模型的位置
2.  DATA  
    2.1 PATH\_TO\_DATA\_DIR: ‘/disk6T/ava’ 这里是ava文件的位置

其他也比较简单，大家就自己理解了。

# 五，训练 #

python tools/run_net.py --cfg configs/AVA/SLOWFAST_32x2_R50_SHORT4.yaml

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 17]

# 六，结果查看 #

这里就要参考之前的博客进行结果的查看：  
[【SlowFast复现】SlowFast Networks for Video Recognition复现代码 使用自己的视频进行demo检测][SlowFast_SlowFast Networks for Video Recognition_ _demo]

使用训练后的权重，对我们的视频进行检测，会发现检测结果很糟糕。毕竟只是用了1个训练视频和1个验证视频

最后检验的视频会上传到B站中。

[Link 1]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_1
[Link 2]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_5
[1.1 myava]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#11_myava_8
[1.2 annotations]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#12_annotations_12
[1.3 annotations]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#13_annotations_88
[1.4 frame_lists]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#14_frame_lists_123
[1.5 frames]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#15_frames_141
[Link 3]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_147
[Link 4]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_155
[4.1 _yaml]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#41_yaml_156
[4.2 yaml]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#42_yaml_258
[Link 5]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_269
[Link 6]: https://blog.csdn.net/WhiffeYF/article/details/115444694?spm=1001.2014.3001.5501#_277
[b]: https://www.bilibili.com/video/BV1eh411U78b?p=1
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70]: /images/20220829/4cf933f8fb1642158fed8420c6fca6d5.png
[Link 7]: https://github.com/facebookresearch/SlowFast/blob/master/slowfast/datasets/DATASET.md
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 1]: /images/20220829/0feb44ea5575428ca8cdf795022f5c92.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 2]: /images/20220829/40224fce6bd74fe7ae09f4b1ad65728e.png
[https_dl.fbaipublicfiles.com_pyslowfast_annotation_ava_ava_annotations.tar]: https://dl.fbaipublicfiles.com/pyslowfast/annotation/ava/ava_annotations.tar
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 3]: /images/20220829/542ac34c9acd442189d8d45e59592617.png
[ava_detection_train_boxes_and_labels_include_negative_v2.2.cs]: /images/20220829/b7619b5c42074883b5c70ddf2cd38649.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 4]: /images/20220829/ea78acfd2aca40d1859a5d3314a3500b.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 5]: /images/20220829/f40d892620a140399579323afb320827.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 6]: /images/20220829/96e38e0846f840189377a50b07a90b74.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 7]: /images/20220829/60077eb70c4246d0a925a3fd6415f6a4.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 8]: /images/20220829/0e923dd909f640299de5432c65c7756d.png
[train.csv]: https://dl.fbaipublicfiles.com/video-long-term-feature-banks/data/ava/frame_lists/train.csv
[val.csv]: https://dl.fbaipublicfiles.com/video-long-term-feature-banks/data/ava/frame_lists/val.csv
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 9]: /images/20220829/9e52d51f6cb94989b5665828755ad0d9.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 10]: /images/20220829/45d119fe699045118c9632d0387de524.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 11]: /images/20220829/d0573db8a13a471899c149e72a7c7af0.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 12]: /images/20220829/dd272097051149228702b045a4fd83c0.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 13]: /images/20220829/bde4a25935434ec6baef96791fb989c4.png
[Link 8]: https://github.com/facebookresearch/SlowFast/blob/master/MODEL_ZOO.md
[Link 9]: https://dl.fbaipublicfiles.com/pyslowfast/model_zoo/ava/pretrain/SLOWFAST_32x2_R101_50_50.pkl
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 14]: /images/20220829/c284c085c0ca4df6a1692765701b7831.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 15]: /images/20220829/404cc785104345498135ca419c0db36c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 16]: /images/20220829/d9b7dc6ec5ce45889380dbb1aa297fc4.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1doaWZmZVlG_size_16_color_FFFFFF_t_70 17]: /images/20220829/11af4996243a473f9a8207a71e0f359e.png
[SlowFast_SlowFast Networks for Video Recognition_ _demo]: https://blog.csdn.net/WhiffeYF/article/details/113527759