Open Access

Table 1

Architectures of the 2D and 3D TW.

(a) 2D TW encoder (E) and decoder (D) specifications.
Input shape 128 × 128
Filter sizes {5, 5, 5, 5, 5}
nfilter(E) {32, 64, 128, 256, 512}
nfilter(D) {256, 128, 64, 32, 1}
Strides {2, 2, 2, 2, 2}
Layer activation Leaky ReLU (E), ReLU (D)
Final activation None(E), Tanh (D)
Latent dimension 100

(b) 3D TW encoder (E) and decoder (D) specifications.
Input shape 64 × 64 × 64
Filter sizes {4, 4, 4, 4}
nfilter(E) {32, 64, 128, 256}
nfilter(D) {128, 64, 32, 1}
Strides {2, 2, 2, 2}
Layer activation Leaky ReLU (E), ReLU (D)
Final activation None(E), Tanh (D)
Latent dimension 200

Notes. All models are trained using the Adam optimizer with the same parameters as above (lr = 0.0002, β1 = 0.5).

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.