search

RQ-Wu / RIDCP_dehazing

[CVPR 2023] | RIDCP: Revitalizing Real Image Dehazing via High-Quality Codebook Priors
https://rq-wu.github.io/projects/RIDCP/index.html
Other
194 stars 23 forks source link

About Model Efficiency #7

Closed koihoo closed 1 year ago

koihoo commented 1 year ago

I did the efficiency test of image dehazing, and the model only got 1.5fps or even lower. Did you test it when doing the experiment? image

RQ-Wu commented 1 year ago

We have not prioritized the efficiency of the model, which is a point that could be further optimized. Can you tell us what the resolution of the image is? Also, as it takes a lot of time to load the model, processing only one image may slow down the average speed

Owen718 commented 1 year ago

It's important to consider the asynchronous nature of the CUDA API and the warm-up time required by your graphics card for optimal performance. For more detailed information on these topics, I recommend consulting both the MIMO-UNet paper and its accompanying code release. @koihoo

koihoo commented 1 year ago

We have not prioritized the efficiency of the model, which is a point that could be further optimized. Can you tell us what the resolution of the image is? Also, as it takes a lot of time to load the model, processing only one image may slow down the average speed

The resolution of the image is 640*438，and I tried to experiment with a whole folder of pictures, including a total of 7 pictures, but the algorithm efficiency is still very low。 image

RQ-Wu commented 1 year ago

We have not prioritized the efficiency of the model, which is a point that could be further optimized. Can you tell us what the resolution of the image is? Also, as it takes a lot of time to load the model, processing only one image may slow down the average speed

The resolution of the image is 640*438，and I tried to experiment with a whole folder of pictures, including a total of 7 pictures, but the algorithm efficiency is still very low。 image

Thanks for your experiments. We do not particularly focus on the inference speed, since we regard dehazing as a computational photography task. How to make RIDCP more efficient can be valuable future work.