FDInSky
FDInSky
I export my model use static graph, but got output is dynamic! 
CenterMask is a one-stage detector, Why did you use two-stage architecture ? Thanks
I see Halide had autodiff func by this paper, https://people.csail.mit.edu/tzumao/gradient_halide/gradient_halide.pdf Will you plan to add autodiff for tiramisu ?
How to add autodiff like Halide? https://people.csail.mit.edu/tzumao/gradient_halide/gradient_halide.pdf
https://github.com/tensor-compiler/taco I want to make taco's tensor computation with auto-diff. Is it possible?
Until now, pytorch have so much good model. Because of static graph, it have to use tensorflow ?