赞
踩
公式:
T
2
(
V
)
=
h
ϕ
(
∑
i
<
j
g
θ
(
f
i
,
f
j
)
)
T_{2}(V)=h_{\phi}\left(\sum_{i<j} g_{\theta}\left(f_{i}, f_{j}\right)\right)
T2(V)=hϕ(i<j∑gθ(fi,fj))
T 3 ( V ) = h ϕ ′ ( ∑ i < j < k g θ ′ ( f i , f j , f k ) ) T_{3}(V)=h_{\phi}^{\prime}\left(\sum_{i<j<k} g_{\theta}^{\prime}\left(f_{i}, f_{j}, f_{k}\right)\right) T3(V)=hϕ′⎝⎛i<j<k∑gθ′(fi,fj,fk)⎠⎞
T 4 ( V ) = h ϕ ′ ( ∑ i < j < k < l g θ ′ ( f i , f j , f k , f l ) ) T_{4}(V)=h_{\phi}^{\prime}\left(\sum_{i<j<k<l} g_{\theta}^{\prime}\left(f_{i}, f_{j}, f_{k}, f_{l}\right)\right) T4(V)=hϕ′⎝⎛i<j<k<l∑gθ′(fi,fj,fk,fl)⎠⎞
M T N ( V ) = T 2 ( V ) + T 3 ( V ) … + T N ( V ) M T_{N}(V)=T_{2}(V)+T_{3}(V) \ldots+T_{N}(V) MTN(V)=T2(V)+T3(V)…+TN(V)
文章指出,TRN的强大之处在于能够解释视频常识,这个是优于C3D和I3D(后续会介绍,常用算法)的地方:
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。