Closed liuxingbin closed 11 months ago
Hi there, great question! You can indeed use attention_FLIS.py for COCO and ADE20K during inference as these lines of code would help us to handle words with more than one token. We opted to apply attention_ADE20K.py and attention_COCO.py, in the sense that this kind of "hard“ coding will save time especially during training.
Feel free to ask if you have any remaining questions.
That's helped me a lot. Thanks for your reply.
Hi, FreestyleNet is a great work. However, I am confused about the attention mechanism. During inference, we use attention_FLIS.py, which involves words with more than one token. Then why we should specifically design the attention for COCO and ADE20k?
I am looking forward to your reply.