a little bean
Results
2
issues of
a little bean
我安装好了pycuda和RT,运行python run_latency_stages.py,也生成了model.onnx,但是1080Ti测速STDC2-Seg75 FPS只有65.8,STDC2-Seg50 FPS只有133.64,和与原文差30-50这么多,不知道RT加速还需要注意什么吗(而且不加速之前用pytorch FPS只有35和58)
Did I not understand the code properly? The cosine similarity is not used in FA, but the dot product is still used. The formula 1/n in the paper is not...