add pose_n_num_encoder

modules/pts_num_encoder.py

@@ -0,0 +1,20 @@
+from torch import nn
+import PytorchBoot.stereotype as stereotype
+
+@stereotype.module("pts_num_encoder")
+class PointsNumEncoder(nn.Module):
+    def __init__(self, config):
+        super(PointsNumEncoder, self).__init__()
+        self.config = config
+        out_dim = config["out_dim"]
+        self.act = nn.ReLU(True)
+
+        self.pts_num_encoder = nn.Sequential(
+            nn.Linear(1, out_dim),
+            self.act,
+            nn.Linear(out_dim, out_dim),
+            self.act,
+        )
+
+    def encode_pts_num(self, num_seq):
+        return self.pts_num_encoder(num_seq)

modules/transformer_pose_n_num_seq_encoder.py

@@ -0,0 +1,72 @@
+import torch
+from torch import nn
+from torch.nn.utils.rnn import pad_sequence
+import PytorchBoot.stereotype as stereotype
+
+
+@stereotype.module("transformer_pose_n_num_seq_encoder")
+class TransformerPoseAndNumSequenceEncoder(nn.Module):
+    def __init__(self, config):
+        super(TransformerPoseAndNumSequenceEncoder, self).__init__()
+        self.config = config
+        embed_dim = config["pts_num_embed_dim"] + config["pose_embed_dim"]
+        encoder_layer = nn.TransformerEncoderLayer(
+            d_model=embed_dim,
+            nhead=config["num_heads"],
+            dim_feedforward=config["ffn_dim"],
+            batch_first=True,
+        )
+        self.transformer_encoder = nn.TransformerEncoder(
+            encoder_layer, num_layers=config["num_layers"]
+        )
+        self.fc = nn.Linear(embed_dim, config["output_dim"])
+
+    def encode_sequence(self, pts_num_embedding_list_batch, pose_embedding_list_batch):
+        combined_features_batch = []
+        lengths = []
+
+        for pts_num_embedding_list, pose_embedding_list in zip(pts_num_embedding_list_batch, pose_embedding_list_batch):
+            combined_features = [
+                torch.cat((pts_num_embed, pose_embed), dim=-1)
+                for pts_num_embed, pose_embed in zip(pts_num_embedding_list, pose_embedding_list)
+            ]
+            combined_features_batch.append(torch.stack(combined_features))
+            lengths.append(len(combined_features))
+
+        combined_tensor = pad_sequence(combined_features_batch, batch_first=True)  # Shape: [batch_size, max_seq_len, embed_dim]
+
+        max_len = max(lengths)
+        padding_mask = torch.tensor([([0] * length + [1] * (max_len - length)) for length in lengths], dtype=torch.bool).to(combined_tensor.device)
+
+        transformer_output = self.transformer_encoder(combined_tensor, src_key_padding_mask=padding_mask)
+        final_feature = transformer_output.mean(dim=1)
+        final_output = self.fc(final_feature)
+
+        return final_output
+
+
+if __name__ == "__main__":
+    config = {
+        "pts_num_embed_dim": 128, 
+        "pose_embed_dim": 256,
+        "num_heads": 4,  
+        "ffn_dim": 256,
+        "num_layers": 3, 
+        "output_dim": 2048, 
+    }
+
+    encoder = TransformerPoseAndNumSequenceEncoder(config)
+    seq_len = [5, 8, 9, 4]
+    batch_size = 4
+
+    pts_num_embedding_list_batch = [
+        torch.randn(seq_len[idx], config["pts_num_embed_dim"]) for idx in range(batch_size)
+    ]
+    pose_embedding_list_batch = [
+        torch.randn(seq_len[idx], config["pose_embed_dim"]) for idx in range(batch_size)
+    ]
+    output_feature = encoder.encode_sequence(
+        pts_num_embedding_list_batch, pose_embedding_list_batch
+    )
+    print("Encoded Feature:", output_feature)
+    print("Feature Shape:", output_feature.shape)

modules/transformer_pose_n_pts_seq_encoder.py

@@ -4,7 +4,7 @@ from torch.nn.utils.rnn import pad_sequence
 import PytorchBoot.stereotype as stereotype
 
 
-@stereotype.module("transformer_seq_encoder")
+@stereotype.module("transformer_pose_n_pts_seq_encoder")
 class TransformerSequenceEncoder(nn.Module):
     def __init__(self, config):
         super(TransformerSequenceEncoder, self).__init__()