videoswin_k400_videos.yaml

MODEL: #MODEL field
    framework: "RecognizerTransformer" #Mandatory, indicate the type of network, associate to the 'paddlevideo/modeling/framework/' .
    backbone: #Mandatory, indicate the type of backbone, associate to the 'paddlevideo/modeling/backbones/' .
        name: "SwinTransformer3D" #Mandatory, The name of backbone.
        pretrained: "data/SwinTransformer_imagenet.pdparams" #Optional, pretrained model path.
        patch_size: [2, 4, 4]
        embed_dim: 128
        depths: [2, 2, 18, 2]
        num_heads: [4, 8, 16, 32]
        window_size: [8, 7, 7]
        mlp_ratio: 4.
        qkv_bias: True
        qk_scale: None
        drop_rate: 0.0
        attn_drop_rate: 0.0
        drop_path_rate: 0.2
        patch_norm: True
    head:
        name: "I3DHead" #Mandatory, indicate the type of head, associate to the 'paddlevideo/modeling/heads'
        num_classes: 400 #Optional, the number of classes to be classified.
        in_channels: 1024 #input channel of the extracted feature.
        spatial_type: 'avg'
        drop_ratio: 0.5 #the ratio of dropout
        std: 0.01 #std value in params initialization
    runtime_cfg: # configuration used when the model is train or test.
        test: # test config
            num_seg: 32
            avg_type: 'prob' # 'score' or 'prob'

DATASET: #DATASET field
    batch_size: 1 #Mandatory, bacth size
    num_workers: 4 #Mandatory, XXX the number of subprocess on each GPU.
    test_batch_size: 1
    train:
        format: "VideoDataset" #Mandatory, indicate the type of dataset, associate to the 'paddlevidel/loader/dateset'
        data_prefix: "data/k400/videos" #Mandatory, train data root path
        file_path: "data/k400/train.list" #Mandatory, train data index file path
    valid:
        format: "VideoDataset" #Mandatory, indicate the type of dataset, associate to the 'paddlevidel/loader/dateset'
        data_prefix: "data/k400/videos" #Mandatory, train data root path
        file_path: "data/k400/val.list" #Mandatory, valid data index file path
    test:
        format: "VideoDataset" #Mandatory, indicate the type of dataset, associate to the 'paddlevidel/loader/dateset'
        data_prefix: "data/k400/videos" #Mandatory, train data root path
        file_path: "data/k400/val.list" #Mandatory, valid data index file path

PIPELINE: #PIPELINE field TODO.....
    train: #Mandotary, indicate the pipeline to deal with the training data, associate to the 'paddlevideo/loader/pipelines/'
        decode:
            name: "VideoDecoder"
            backend: 'decord'
            mode: 'train'
        sample:
            name: "Sampler"
            num_seg: 1
            frame_interval: 2
            seg_len: 32
            valid_mode: False
            use_pil: False
        transform: #Mandotary, image transform operator.
            - Scale:
                short_size: 256
                fixed_ratio: False
                keep_ratio: True
                backend: 'cv2'
                do_round: True
            - RandomResizedCrop:
                backend: 'cv2'
            - Scale:
                short_size: 224
                fixed_ratio: False
                keep_ratio: False
                backend: 'cv2'
                do_round: True
            - RandomFlip:
            - Normalization:
                mean: [123.675, 116.28, 103.53]
                std: [58.395, 57.12, 57.375]
                tensor_shape: [3, 1, 1, 1]
                inplace: True
            - Image2Array:
                data_format: 'cthw'
    valid: #Mandatory, indicate the pipeline to deal with the validing data. associate to the 'paddlevideo/loader/pipelines/'
        decode:
            name: "VideoDecoder"
            backend: 'decord'
            mode: 'valid'
        sample:
            name: "Sampler"
            num_seg: 1
            frame_interval: 2
            seg_len: 32
            valid_mode: True
            use_pil: False
        transform: #Mandotary, image transform operator.
            - Scale:
                short_size: 256
                fixed_ratio: False
                keep_ratio: True
                backend: 'cv2'
                do_round: True
            - CenterCrop:
                target_size: 224
                do_round: False
                backend: 'cv2'
            - Normalization:
                mean: [123.675, 116.28, 103.53]
                std: [58.395, 57.12, 57.375]
                tensor_shape: [3, 1, 1, 1]
                inplace: True
            - Image2Array:
                data_format: 'cthw'
    test:
        decode:
            name: "VideoDecoder"
            backend: 'decord'
            mode: 'valid'
        sample:
            name: "Sampler"
            num_seg: 4
            frame_interval: 2
            seg_len: 32
            valid_mode: True
            use_pil: False
        transform: #Mandotary, image transform operator.
            - Scale:
                short_size: 224
                fixed_ratio: False
                keep_ratio: True
                backend: 'cv2'
                do_round: True
            - UniformCrop:
                target_size: 224
                backend: 'cv2'
            - Normalization:
                mean: [123.675, 116.28, 103.53]
                std: [58.395, 57.12, 57.375]
                tensor_shape: [3, 1, 1, 1]
                inplace: True
            - Image2Array:
                data_format: 'cthw'

OPTIMIZER: #OPTIMIZER field
    name: 'AdamW' #Mandatory, the type of optimizer, associate to the 'paddlevideo/solver/'
    beta1: 0.9
    beta2: 0.999
    no_weight_decay_name: 'norm relative_position_bias_table'
    learning_rate: #Mandatory, the type of learning rate scheduler, associate to the 'paddlevideo/solver/'
        name: 'CustomWarmupCosineStepDecay'
        iter_step: True
        warmup_iters: 2.5
        warmup_ratio: 0.1
        min_lr: 0
        base_lr: 3e-5
        max_epoch: 30
    weight_decay: 0.05

METRIC:
    name: 'CenterCropMetric'

GRADIENT_ACCUMULATION:
    global_batch_size: 64 # Specify the sum of batches to be calculated by all GPUs

INFERENCE:
    name: 'VideoSwin_Inference_helper'
    num_seg: 1
    seg_len: 32
    short_size: 256
    target_size: 224

model_name: "VideoSwin"
log_interval: 20 #Optional, the interal of logger, default:10
save_interval: 5
epochs: 30 #Mandatory, total epoch
log_level: "INFO" #Optional, the logger level. default: "INFO"