video-swin-transformer/video_swin_transformer.py

import logging
import os
import json
from pathlib import Path
from typing import List
import torch
import numpy
from towhee import register
from towhee.operator.base import NNOperator
from towhee.types.video_frame import VideoFrame
from towhee.models.utils.video_transforms import transform_video, get_configs
from towhee.models.video_swin_transformer import video_swin_transformer
from towhee.models.video_swin_transformer.get_configs import configs
log = logging.getLogger()


@register(output_schema=['labels', 'scores', 'features'])
class VideoSwinTransformer(NNOperator):
    """
    Generate a list of class labels given a video input data.
    Default labels are from [Kinetics400 Dataset](https://deepmind.com/research/open-source/kinetics).

    Args:
        model_name (`str`):
            Supported model names:
            - swin_t_w877_k400_1k
        skip_preprocess (`str`):
            Flag to skip video transforms.
        classmap (`str=None`):
            Path of the json file to match class names.
        topk (`int=5`):
            The number of classification labels to be returned (ordered by possibility from high to low).
    """
    def __init__(self,
                 model_name: str = 'swin_t_w877_k400_1k',
                 framework: str = 'pytorch',
                 skip_preprocess: bool = False,
                 classmap: str = None,
                 topk: int = 5,
                 ):
        super().__init__(framework=framework)
        self.model_name = model_name
        self.skip_preprocess = skip_preprocess
        self.topk = topk
        self.model_configs = configs(self.model_name)
        if classmap is None:
            class_file = os.path.join(str(Path(__file__).parent), self.model_configs['labels_file_name'])
            with open(class_file, 'r') as f:
                kinetics_classes = json.load(f)
            self.classmap = {}
            for k, v in kinetics_classes.items():
                self.classmap[v] = str(k).replace('"', '')
        else:
            self.classmap = classmap
        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'

        self.model = video_swin_transformer.create_model(model_name=self.model_name,
                                                         pretrained=True,
                                                         device=self.device)

        self.transform_cfgs = get_configs(
            side_size=224,
            crop_size=224,
            num_frames=32,
            mean=[123.675, 116.28, 103.53],
            std=[58.395, 57.12, 57.375],
        )

    def decoder_video(self,  data: List[VideoFrame]):
        video = numpy.stack([img.astype(numpy.float32) / 255. for img in data], axis=0)
        assert len(video.shape) == 4
        video = video.transpose(3, 0, 1, 2)  # twhc -> ctwh
        video = transform_video(
            video=video,
            **self.transform_cfgs
        )
        # [B x C x T x H x W]
        video = video.to(self.device)[None, ...]
        return video

    def __call__(self, video: List[VideoFrame]):
        """
        Args:
            video (`List[VideoFrame]`):
                Video path in string.

        Returns:
            (labels, scores)
                A tuple of lists (labels, scores).
            OR emb
                Video embedding.
        """
        inputs = self.decoder_video(video)
        # inputs [B x C x T x H x W]
        feats = self.model.forward_features(inputs)
        features = feats.to('cpu').squeeze(0).detach().numpy()

        outs = self.model.head(feats)
        post_act = torch.nn.Softmax(dim=1)
        preds = post_act(outs)
        pred_scores, pred_classes = preds.topk(k=self.topk)
        labels = [self.classmap[int(i)] for i in pred_classes[0]]
        scores = [round(float(x), 5) for x in pred_scores[0]]

        return labels, scores, features