| 101 | |
| 102 | |
| 103 | class MLDecoder(nn.Module): |
| 104 | def __init__(self, num_classes, num_of_groups=-1, decoder_embedding=768, initial_num_features=2048): |
| 105 | super(MLDecoder, self).__init__() |
| 106 | embed_len_decoder = 100 if num_of_groups < 0 else num_of_groups |
| 107 | if embed_len_decoder > num_classes: |
| 108 | embed_len_decoder = num_classes |
| 109 | |
| 110 | # switching to 768 initial embeddings |
| 111 | decoder_embedding = 768 if decoder_embedding < 0 else decoder_embedding |
| 112 | self.embed_standart = nn.Linear(initial_num_features, decoder_embedding) |
| 113 | |
| 114 | # decoder |
| 115 | decoder_dropout = 0.1 |
| 116 | num_layers_decoder = 1 |
| 117 | dim_feedforward = 2048 |
| 118 | layer_decode = TransformerDecoderLayerOptimal(d_model=decoder_embedding, |
| 119 | dim_feedforward=dim_feedforward, dropout=decoder_dropout) |
| 120 | self.decoder = nn.TransformerDecoder(layer_decode, num_layers=num_layers_decoder) |
| 121 | |
| 122 | # non-learnable queries |
| 123 | self.query_embed = nn.Embedding(embed_len_decoder, decoder_embedding) |
| 124 | self.query_embed.requires_grad_(False) |
| 125 | |
| 126 | # group fully-connected |
| 127 | self.num_classes = num_classes |
| 128 | self.duplicate_factor = int(num_classes / embed_len_decoder + 0.999) |
| 129 | self.duplicate_pooling = torch.nn.Parameter( |
| 130 | torch.Tensor(embed_len_decoder, decoder_embedding, self.duplicate_factor)) |
| 131 | self.duplicate_pooling_bias = torch.nn.Parameter(torch.Tensor(num_classes)) |
| 132 | torch.nn.init.xavier_normal_(self.duplicate_pooling) |
| 133 | torch.nn.init.constant_(self.duplicate_pooling_bias, 0) |
| 134 | self.group_fc = GroupFC(embed_len_decoder) |
| 135 | |
| 136 | def forward(self, x): |
| 137 | if len(x.shape) == 4: # [bs,2048, 7,7] |
| 138 | embedding_spatial = x.flatten(2).transpose(1, 2) |
| 139 | else: # [bs, 197,468] |
| 140 | embedding_spatial = x |
| 141 | embedding_spatial_786 = self.embed_standart(embedding_spatial) |
| 142 | embedding_spatial_786 = torch.nn.functional.relu(embedding_spatial_786, inplace=True) |
| 143 | |
| 144 | bs = embedding_spatial_786.shape[0] |
| 145 | query_embed = self.query_embed.weight |
| 146 | # tgt = query_embed.unsqueeze(1).repeat(1, bs, 1) |
| 147 | tgt = query_embed.unsqueeze(1).expand(-1, bs, -1) # no allocation of memory with expand |
| 148 | h = self.decoder(tgt, embedding_spatial_786.transpose(0, 1)) # [embed_len_decoder, batch, 768] |
| 149 | h = h.transpose(0, 1) |
| 150 | |
| 151 | out_extrap = torch.zeros(h.shape[0], h.shape[1], self.duplicate_factor, device=h.device, dtype=h.dtype) |
| 152 | self.group_fc(h, self.duplicate_pooling, out_extrap) |
| 153 | h_out = out_extrap.flatten(1)[:, :self.num_classes] |
| 154 | h_out += self.duplicate_pooling_bias |
| 155 | logits = h_out |
| 156 | return logits |
no outgoing calls
no test coverage detected