search

Vincent-ZHQ / UniS-MMC

Code for UniS-MMC: Multimodal Classification via Unimodality-supervised Multimodal Contrastive Learning (ACL 2023)
MIT License
28 stars 4 forks source link

The question of Pre-training #4

Open fishermanfff opened 6 months ago

fishermanfff commented 6 months ago

In your paper, the accuracy of the food data set reached 94%. What version of bert and vit did you use? When I reproduced your code, I found that you only provided the base version.

lu-student commented 6 months ago

Hello, dataloader.py part, where is the file in the /text_txt/ folder in this code text_path = self.args.data_dir + '/texts_txt/' + annotation + '/' + id.replace(".jpg", ".txt").

Vincent-ZHQ commented 6 months ago

In your paper, the accuracy of the food data set reached 94%. What version of bert and vit did you use? When I reproduced your code, I found that you only provided the base version.

The results are all from models with BERT-base and ViT-base. I also test it with the large version. It could provide better results with larger GPU memory.