diff options
author | Jordan Gong <jordan.gong@protonmail.com> | 2021-04-06 21:19:26 +0800 |
---|---|---|
committer | Jordan Gong <jordan.gong@protonmail.com> | 2021-04-06 21:19:26 +0800 |
commit | 7f6c65fa954a43c9d248219525cad35fe1b8a046 (patch) | |
tree | 67fb188dd59b517a02ac25cea630e9f094522afb /config.py | |
parent | b9f35fbe7d78b3c478086ea26c2a76f72ce35687 (diff) |
Network modification
This commit change the auto-encoder by removing fc and optimizing latent space features
Diffstat (limited to 'config.py')
-rw-r--r-- | config.py | 4 |
1 files changed, 2 insertions, 2 deletions
@@ -11,7 +11,7 @@ config: Configuration = { # Recorde disentangled image or not 'image_log_on': False, # The number of subjects for validating (Part of testing set) - 'val_size': 10, + 'val_size': 20, }, # Dataset settings 'dataset': { @@ -51,7 +51,7 @@ config: Configuration = { # Auto-encoder feature channels coefficient 'ae_feature_channels': 64, # Appearance, canonical and pose feature dimensions - 'f_a_c_p_dims': (192, 192, 96), + 'f_a_c_p_dims': (192, 192, 128), # HPM pyramid scales, of which sum is number of parts 'hpm_scales': (1, 2, 4, 8), # Global pooling method |