jokieleung / lpf-vqa

the code for SIGIR 2021 "LPF: A Language-Prior Feedback Objective Function for De-biased Visual Question Answering"

Hi 👋, I'm Zujie Liang

Welcome! You are the visitor!

• 😄 Hi there, this is Zujie Liang (or "Jokie Leung" in Cantonese). I am a master student at Sun Yat-Sen University (SYSU).
• 🔭 My research interest includes Natural Language Processing, Machine Learning, and Computer Vision.
• 👯 I am open to collaboration on research, feel free to contact me via E-mail!
• ⚡ Please check my homepage for my CV and latest update!

lpf-vqa's Issues

updw +LPF: mat1 and mat2 shapes cannot be multiplied (24588x1024 and 4800x1)

Dear scholar:

When I run python -m bootstrap.run -o rubi/options/vqacp2/lpf_updn.yaml
The code throw exceptions as follows:
LPF-VQA/rubi/models/networks/updn.py", line 90, in logits logits = self.linear(joint_repr)
RuntimeError: mat1 and mat2 shapes cannot be multiplied (24588x1024 and 4800x1)

我直接将q_dim 改为1024 即可跑通
class NewAttention(nn.Module):
def init(self, v_dim, q_dim, num_hid, dropout=0.2):
super(NewAttention, self).init()

    self.v_proj = FCNet([v_dim, num_hid])
    self.q_proj = FCNet([q_dim, num_hid])
    self.dropout = nn.Dropout(dropout )
    #self.linear = weight_norm(nn.Linear(q_dim, 1), dim=None)
    self.linear = weight_norm(nn.Linear(1024, 1), dim=None)