Table 2 ConvLSTM baseline architecture detailed description
From: A novel human activity recognition architecture: using residual inception ConvLSTM layer
Layer ID | Layer name | Input size | Output size | Number of parameters | Input ID | Return sequence |
|---|---|---|---|---|---|---|
system_input | Input | (16, 20, 40, 40, 1) | (16, 20, 40, 40, 1) | 0 | - | - |
convlstm_1_3x3 | ConvLSTM | (16, 20, 40, 40, 1) | (16, 20, 40, 40, 40) | 59,200 | system_input | Yes |
batch_norm_1 | Batch Normalization | (16, 20, 40, 40, 40) | (16, 20, 40, 40, 40) | 160 | convlstm_1_3x3 | - |
convlstm_2_3x3 | ConvLSTM | (16, 20, 40, 40, 40) | (16, 40, 40, 1) | 1480 | batch_norm_1 | No |
batch_norm_2 | Batch Normalization | (16, 40, 40, 1) | (16, 40, 40, 1) | 4 | convlstm_2_3x3 | - |
reshape | Reshape | (16, 40, 40, 1) | (16, 1, 1600) | 0 | batch_norm_2 | - |
dense_1 | Dense | (16, 1, 1600) | (16, 1, 6) | 9606 | reshape | - |
dropout | Dropout | (16, 1, 6) | (16, 1, 6) | 0 | dense_1 | - |
dense_1 | Dense | (16, 1, 6) | (16, 1, 6) | 42 | dropout | - |