Based on PyTorch
Built using PyTorch. Makes it easy to use all the PyTorch-ecosystem components.
Reproducible Model Zoo
Variety of state of the art pretrained video models and their associated benchmarks that are ready to use.
Efficient Video Components
Video-focused fast and efficient components that are easy to use. Supports accelerated inference on hardware.
pip install pytorchvideo
# Import all the required components
...
# Load pre-trained model
model = torch.hub.load('facebookresearch/pytorchvideo', 'slow_r50', pretrained=True)
# Load video
video = EncodedVideo.from_path('some_video.avi')
# Compose video data transforms
transform = ApplyTransformToKey(
key="video",
transform=Compose(
[
UniformTemporalSubsample(num_frames),
Lambda(lambda x: x/255.0),
NormalizeVideo(mean, std),
ShortSideScale(
size=side_size
),
CenterCropVideo(crop_size=(crop_size, crop_size))
]
),
)
# Get clip
clip_start_sec = 0.0 # secs
clip_duration = 2.0 # secs
video_data = video.get_clip(start_sec=clip_start_sec, end_sec=clip_start_sec + clip_duration)
video_data = transform(video_data)
# Generate top 5 predictions
preds = torch.nn.functional.softmax(preds)
pred_class_ids = preds.topk(k=5).indices