| import torch |
| import torch.nn as nn |
|
|
| class ResBlock(nn.Module): |
| def __init__(self, dim, dropout=0.3): |
| super(ResBlock, self).__init__() |
| self.block = nn.Sequential( |
| nn.Linear(dim, dim), |
| nn.BatchNorm1d(dim), |
| nn.GELU(), |
| nn.Dropout(dropout), |
| nn.Linear(dim, dim), |
| nn.BatchNorm1d(dim), |
| nn.GELU(), |
| nn.Dropout(dropout) |
| ) |
| def forward(self, x): |
| return x + self.block(x) |
|
|
| class Yoga3HeadMLP(nn.Module): |
| def __init__(self, input_dim, num_poses, num_joints=15): |
| super(Yoga3HeadMLP, self).__init__() |
| self.input_layer = nn.Sequential( |
| nn.Linear(input_dim, 256), |
| nn.BatchNorm1d(256), |
| nn.GELU() |
| ) |
| self.res1 = ResBlock(256, dropout=0.3) |
| self.res2 = ResBlock(256, dropout=0.3) |
| |
| |
| self.pose_head = nn.Sequential( |
| nn.Linear(256, 128), |
| nn.BatchNorm1d(128), |
| nn.GELU(), |
| nn.Dropout(0.2), |
| nn.Linear(128, num_poses) |
| ) |
| |
| |
| self.correctness_head = nn.Sequential( |
| nn.Linear(256, 64), |
| nn.BatchNorm1d(64), |
| nn.GELU(), |
| nn.Dropout(0.2), |
| nn.Linear(64, 1) |
| ) |
| |
| |
| self.deviation_head = nn.Sequential( |
| nn.Linear(256, 128), |
| nn.BatchNorm1d(128), |
| nn.GELU(), |
| nn.Dropout(0.2), |
| nn.Linear(128, num_joints) |
| ) |
| |
| def forward(self, x): |
| features = self.input_layer(x) |
| features = self.res1(features) |
| features = self.res2(features) |
| |
| pose_logits = self.pose_head(features) |
| correctness_logit = self.correctness_head(features).squeeze(-1) |
| deviations = self.deviation_head(features) |
| |
| return pose_logits, correctness_logit, deviations |
|
|
