e2e_thumos_videomaev2_g_768x1_160_adapter.py

_base_ = ["e2e_thumos_videomae_s_768x1_160_adapter.py"]

model = dict(
    backbone=dict(
        backbone=dict(
            patch_size=14,
            embed_dims=1408,
            depth=40,
            num_heads=16,
            mlp_ratio=48 / 11,
            adapter_index=list(range(40)),
        ),
        custom=dict(pretrain="pretrained/vit-giant-p14_videomaev2-hybrid_pt_1200e_k710_ft_my.pth"),
    ),
    projection=dict(in_channels=1408),
)

optimizer = dict(backbone=dict(custom=[dict(name="adapter", lr=1e-4, weight_decay=0.05)]))

work_dir = "exps/thumos/adatad/e2e_actionformer_videomaev2_g_768x1_160_adapter"