search

SooLab / CoTDet

[ICCV2023] CoTDet: Affordance Knowledge Prompting for Task Driven Object Detection
Apache License 2.0
10 stars 0 forks source link

How to generate the knowledge_base.pkl file? #4

Open chaohuchao opened 4 months ago

chaohuchao commented 4 months ago

Thank you for your excellent work! The code CoT_with_ChatGPT.py does not provide the knowledge_base.pkl saving code, but only provides the text output content. Can you provide the specific generation code?

Toneyaya commented 4 months ago

You can easily achieve this using the RoBERTa model from the Transformers library. If you don't consider filtering, use RoBERTa to encode each visual affordance you obtain from MLCoT, utilizing the output sentence embeddings and word embeddings as the key and value for each knowledge unit, respectively. I will update the code as soon as I meet my current deadline. Thank you for your attention.