uncbiag / SimpleClick

SimpleClick: Interactive Image Segmentation with Simple Vision Transformers (ICCV 2023)
MIT License
209 stars 32 forks source link

inference performance tunning #8

Closed xbkaishui closed 1 year ago

xbkaishui commented 1 year ago

hello, thanks for you great work. I want to enhance the model inference performance, can you give some advices. below is some ideas I want to try:

is that ok ?

qinliuliuqin commented 1 year ago

Yes, if you want to improve the speed/memory of SimpleClick, you can use a smaller input size (e.g., 224x224). Using FP16 should also be helpful.

xbkaishui commented 1 year ago

ok. so how to find the smaller model? do you mean just change the input size