Open yu-changqian opened 6 years ago
@ycszen Image resolutions are mentioned in the paper and you can find the corresponding FLOPS using this repository.
I have read this paper. However, it just mentioned the input image resolution. It didn't mention the upsampled image resolution, which influence the performance heavily. In my experiment, my implementation has to upsample the prediction to half of the input image, which can get comparable run-time.
@ycszen in our code input and target have same resolution. eg. for cityscapes input and target resolution is 1024x512
.
Really? I find the speed of my re-implementation is far from the result of the paper. I am doubting why the LinkNet or ENet can run so fast.
Hi, I am reproducing the LinkNet. I have a doubt about the input image resolution and the output image resolution when you compute the FLOPS. I find my FLOPS and running speed are different your results reported on your paper.