search

open-mmlab / mmaction

An open-source toolbox for action understanding based on PyTorch
https://open-mmlab.github.io/
Apache License 2.0
1.86k stars 351 forks source link

Is it possible to use EfficientNet backbone in TSN? #242

Open alejopaullier96 opened 2 years ago

alejopaullier96 commented 2 years ago

Is it possible to use an EfficientNet backbone in TSN? I mean something like this:

model = dict(
    type='Recognizer2D',
    backbone=dict(
        type='EfficientNetB4',
        pretrained='torchvision://efficientnet_b4', <------------- somehow change this
        depth=50,
        norm_eval=False),
    cls_head=dict(
        type='TSNHead',
        num_classes=3,
        in_channels=2048,
        spatial_type='avg',
        consensus=dict(type='AvgConsensus', dim=1),
        dropout_ratio=0.4,
        init_std=0.01),
    train_cfg=None,
    test_cfg=dict(average_clips=None))

I dont know if this is possible. For sure there are no weights in the model zoo for this but maybe if this can be done it can be initialized with PyTorch EfficientNet's weights.

hukkai commented 2 years ago

@alejopaullier96 mmaction is no longer maintained. Please check mmaction2