BREAKPOINT: first attremp based on brepgen encoder and deepsef decoder

brep2sdf/config/default_config.py

@@ -7,11 +7,11 @@ class ModelConfig:
     brep_feature_dim: int = 16
     use_cf: bool = True
     embed_dim: int = 768 # 3 的 倍数
-    latent_dim: int = 16
+    latent_dim: int = 32
     
     # 点云采样配置
-    num_surf_points: int = 8  # 每个面采样点数
+    num_surf_points: int = 16  # 每个面采样点数
-    num_edge_points: int = 2   # 每条边采样点数
+    num_edge_points: int = 4   # 每条边采样点数
     
     # Transformer相关配置
     num_transformer_layers: int = 4
@@ -27,8 +27,8 @@ class ModelConfig:
 @dataclass
 class DataConfig:
     """数据相关配置"""
-    max_face: int = 32
+    max_face: int = 16
-    max_edge: int = 128
+    max_edge: int = 64
     num_query_points: int = 32*32*32  # 限制查询点数量，sdf 采样点数 本来是 128*128*128 ，在data load时随机采样
     bbox_scaled: float = 1.0
 
@@ -57,9 +57,9 @@ class TrainConfig:
     # 基本训练参数
     batch_size: int = 8
     num_workers: int = 4
-    num_epochs: int = 100
+    num_epochs: int = 500
-    learning_rate: float = 1
+    learning_rate: float = 0.01
-    min_lr: float = 1e-1
+    min_lr: float = 1e-5
     weight_decay: float = 0.01
     
     # 梯度和损失相关

brep2sdf/data/data.py

@@ -88,6 +88,8 @@ class BRepSDFDataset(Dataset):
             idx for idx in range(len(self.brep_data_list))
             if (self._get_brep_face_and_edge(self.brep_data_list[idx]) <= (self.max_face, self.max_edge))
         ]
+
+        #filtered_indices = filtered_indices[0:8] # TODO rm
         
         # Use filtered_indices to update brep_data_list and sdf_data_list
         self.brep_data_list = [self.brep_data_list[idx] for idx in filtered_indices]

brep2sdf/networks/loss.py

@@ -16,6 +16,7 @@ class Brep2SDFLoss(nn.Module):
         self.warmup_epochs = warmup_epochs
         
         self.l1_loss = nn.L1Loss(reduction='mean')
+        self.mse_loss = nn.MSELoss(reduction='mean')
     
     def forward(self, pred_sdf, gt_sdf, points=None, epoch=None):
         """计算SDF预测的损失
@@ -33,6 +34,7 @@ class Brep2SDFLoss(nn.Module):
         
         # 2. 计算基础L1损失
         l1_loss = self.l1_loss(pred_sdf, gt_sdf)
+        #mse_loss = self.mse_loss(pred_sdf, gt_sdf)
         
         # 3. 计算梯度损失（如果提供了points）
         grad_loss = 0

brep2sdf/train.py

@@ -29,6 +29,8 @@ def main():
                     init_timeout=180,  # 增加超时时间
                     _disable_stats=True,  # 禁用统计
                     _disable_meta=True,  # 禁用元数据
+                    #http_proxy='https://sub.toline.link/s/fch9w',
+                    #https_proxy='https://sub.toline.link/s/fch9w'
                 ),
                 mode="offline"  # 使用离线模式
             )