def __init__(self, time_steps: int, spatial_encoder_planes: Tuple[int,
...],
bottleneck_planes: int, spatial_decoder_planes: Tuple[int,
...],
classifier_drop_rate: float, class_embed_planes: int,
num_classes: int, vote_type: str):
super(VAETimeAlignedResNet, self).__init__()
self.spatial_encoder_planes = spatial_encoder_planes
self.spatial_bottleneck_planes = bottleneck_planes
self.spatial_decoder_planes = spatial_decoder_planes
self.time_steps = time_steps
self.drop_rate = classifier_drop_rate
self.num_classes = num_classes
self.class_embed_planes = class_embed_planes
self.vote_type = vote_type
self.class_in_planes = self.spatial_bottleneck_planes * self.time_steps
self.spatial_encoder = se.VarSpatialResNetEncoder(
self.spatial_encoder_planes, self.spatial_bottleneck_planes)
self.temporal_encoder = te.VarTemporalResNetEncoder(
self.time_steps, self.spatial_bottleneck_planes)
self.decoder = sd.VarSpatialResNetDecoder(
self.spatial_bottleneck_planes, self.spatial_decoder_planes)
self.classifier = cls.VarTimeAlignedResNetClassifier(
self.class_in_planes, self.class_embed_planes, self.drop_rate,
self.num_classes)
self.softmax = nn.Softmax(dim=-1)
hp.he_init(self)
def __init__(self, time_steps: int, spatial_encoder_planes: Tuple[int,
...],
bottleneck_planes: int, spatial_decoder_planes: Tuple[int,
...],
classifier_drop_rate: float, class_embed_planes: int,
num_classes: int, flow: bool):
super(AETimeAlignedResNet, self).__init__()
self.spatial_encoder_planes = spatial_encoder_planes
self.spatial_bottleneck_planes = bottleneck_planes
self.spatial_decoder_planes = spatial_decoder_planes
self.time_steps = time_steps
self.drop_rate = classifier_drop_rate
self.class_embed_planes = class_embed_planes
self.num_classes = num_classes
self.flow = flow
self.class_in_planes = self.spatial_bottleneck_planes * self.time_steps
self.spatial_encoder = se.SpatialResNetEncoder(
self.spatial_encoder_planes, self.spatial_bottleneck_planes)
self.temporal_encoder = te.TemporalResNetEncoder(
self.time_steps, self.spatial_bottleneck_planes)
self.decoder = sd.SpatialResNetDecoder(self.spatial_bottleneck_planes,
self.spatial_decoder_planes,
self.flow)
self.classifier = cls.TimeAlignedResNetClassifier(
self.class_in_planes, self.class_embed_planes, self.drop_rate,
self.num_classes)
hp.he_init(self)
def __init__(self,
time_steps: int,
num_classes: int,
dropout_prob: float = 0.0):
super(I3D, self).__init__()
self.time_steps = time_steps
self.num_classes = num_classes
self.dropout_prob = dropout_prob
self.encoder = en.I3DEncoder(None)
self.classifier = cls.I3DClassifier(1024, dropout_prob, num_classes)
hp.he_init(self)
def __init__(self, time_steps: int, embed_planes: int, dropout_prob: float,
num_classes: int, flow: bool):
super(AEI3D, self).__init__()
self.time_steps = time_steps
self.num_classes = num_classes
self.dropout_prob = dropout_prob
self.embed_planes = embed_planes
self.flow = flow
self.encoder = en.I3DEncoder(self.embed_planes, flow=self.flow)
self.decoder = de.I3DDecoder(self.embed_planes, self.time_steps,
self.flow)
self.classifier = cls.I3DClassifier(self.embed_planes,
self.dropout_prob,
self.num_classes)
hp.he_init(self)