models/CoCoTask_Model.py

import torch
from .Transformer import TransformerDecoderLayer
from .Transformer import TransformerDecoder
from .ScoreFunction import ScoreFunction

class CoCoTask_Model(torch.nn.Module):
    def __init__(self, 
                 num_layers=3, 
                 norm=None, 
                 return_intermediate=False,
                 d_model = 1024, 
                 nhead = 8, 
                 dim_feedforward=2048, 
                 dropout=0.1,
                 N_words=10, 
                 activation="relu", 
                 normalize_before=False,
                 device = "cuda:0") -> None:
        super().__init__()
        self.decoder_norm = torch.nn.LayerNorm(d_model)
        self.decoder_layer = TransformerDecoderLayer(d_model, nhead, dim_feedforward, dropout, activation, normalize_before)
        self.decoder = TransformerDecoder(self.decoder_layer, num_layers, self.decoder_norm, return_intermediate=return_intermediate)
        self.ScoreFunction = ScoreFunction(N_words=N_words)

        self.threshold = 0.35
    
    def forward(self, tgt, memory, norm=False):
        tgt_new, memory_new = self.decoder(tgt,memory,None)
        score_raw = torch.mm(tgt_new,memory_new.T)
        #NOTE: TODO check later to see if imagebind need softmax
        if norm:
            score_raw = torch.softmax(score_raw)
        #score_res = self.MLP(score_raw)
        score_res = self.ScoreFunction(score_raw)
        return tgt_new, memory_new, score_res