I am trying to test the speed of your models, but I get different FPS compared to yours:
For RCF-VGG16: mine (39 FPS) compared to yours (30 FPS).
For RCF-ResNet50: mine (14 FPS) compared to yours (20 FPS).
For RCF-ResNet101: mine (9.4 FPS) compared to yours (12.2 FPS).
Did you use only one NVIDIA TITAN X GPU (12Gb) for your benchmark? In my benchmark, I used GeForce GTX 1080 Ti (12Gb), but I get more FPS with RCF-VGG16 and less FPS with RCF-ResNet50, RCF-ResNet101 than yours. Moreover, I cannot test RCF-ResNet101-MS because of memory limitation (12Gb). In fact, the images become bigger with scale of 1.5 (for ex, 481x321 becomes 722x482). When I tested with my GPU (12Gb), there was always an error of Cuda: 'error == cudaSuccess (2 vs. 0) out of memory'.
Hello,
I am trying to test the speed of your models, but I get different FPS compared to yours:
Did you use only one NVIDIA TITAN X GPU (12Gb) for your benchmark? In my benchmark, I used GeForce GTX 1080 Ti (12Gb), but I get more FPS with RCF-VGG16 and less FPS with RCF-ResNet50, RCF-ResNet101 than yours. Moreover, I cannot test RCF-ResNet101-MS because of memory limitation (12Gb). In fact, the images become bigger with scale of 1.5 (for ex, 481x321 becomes 722x482). When I tested with my GPU (12Gb), there was always an error of Cuda: 'error == cudaSuccess (2 vs. 0) out of memory'.
Thank you very much.