Changing Encoder and Tokenizer is possible?

IDEA-Research / GroundingDINO

[ECCV 2024] Official implementation of the paper "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection"

https://arxiv.org/abs/2303.05499

Apache License 2.0

6.83k stars 693 forks source link

Changing Encoder and Tokenizer is possible? #287

Open dilicat0813 opened 9 months ago

dilicat0813 commented 9 months ago

HI. Is it possible to change bert-base-uncased to bert-base-multilingual-uncased?

I'v tried to change the code at util, get_tokenizer.py but it didn't work.

rentainhe commented 9 months ago

It may be not useful because we trained BERT for aligning the word and region feature, simply replacing the language model and tokenizer may need to retrain the model