search

TIMMY-CHAN / MISS

[ICANN 2024 (Oral)] MISS: A Generative Pre-training and Fine-tuning Approach for Med-VQA
7 stars 0 forks source link

experiment results #2

Open Victory0622 opened 2 months ago

Victory0622 commented 2 months ago

This is really great work! I ran it on the RAD-VQA dataset without pre-training and obtained lower accuracy rates: 39.1% for open questions, 65.4% for closed questions, and 54.9% overall. I would like to know if these results are right.

TIMMY-CHAN commented 2 months ago

是的，未经预训练的模型直接在下游任务数据上微调将会导致较为糟糕的结果