search

BrandonHanx / mmf

[ECCV 2022] FashionViL: Fashion-Focused V+L Representation Learning
https://mmf.sh/
Other
58 stars 7 forks source link

Run OCIR prediction #4

Closed pntt3011 closed 2 years ago

pntt3011 commented 2 years ago

❓ Questions and Help

Hi, suppose I have PolyvoreOutfits dataset and a list of image ids (like questions attr in ocir_disjoint_test.json), how can I predict the blank id for OCIR task?

I read your evaluation code but it just calculates the similarity between the prediction and ground-truth embedding (please correct me if I'm wrong).

I also run the prediction.py instead of run.py but the report file is just []. Any advice would be much appreciated.

BrandonHanx commented 2 years ago

Yes, you are right, OCIR is a retrieval task rather than a prediction task. We are following the setting of Fashion outfit complementary item retrieval, CVPR2020.

prediction.py is used to predict data without ground truth, you need to define post-processing to get the expected output. We didn't use it in our pipeline.

pntt3011 commented 2 years ago

Thank you for your reply. I will close this now.