From: Fused behavior recognition model based on attention mechanism
ResNet34 | 3D Res18 | ||||
---|---|---|---|---|---|
Layer name | Output size | The architecture of ResNet34 | Layer name | Output size | The architecture of 3D Res18 |
Conv1 | 112 × 112 | [2D conv7 × 7 64] | Conv1 | 7 × 7 × 128 | {3Dconv3 × 3 × 3128 3Dconv3 × 3 × 3128} × 2 |
pool | 56 × 56 | [max pool 3 × 3] | Conv2 | 7 × 7 × 256 | {3Dconv3 × 3 × 3256 3Dconv3 × 3 × 3256} × 2 |
Layer1 | 56 × 56 | {2D conv3 × 3 64 2D conv3 × 3 64} × 3 | Conv3 | 7 × 7 × 512 | {3Dconv3 × 3 × 3512 3Dconv3 × 3 × 3512} × 2 |
Layer2 | 28 × 28 | {2D conv3 × 3128 2D conv3 × 3128} × 4 | Pooling | 1 × 1 × 512 | [Avgpool3D 1 × 7 × 7] |
Layer3 | 14 × 14 | {2D conv3 × 3256 2D conv3 × 3256} × 6 | Dropout | 1 × 8192 | Dropout (p = 0.5) |
Layer4 | 7 × 7 | {2D conv3 × 3512 2D conv3 × 3512} × 3 | – | 1 × classes | FC, softmax |