LiUzHiAn
LiUzHiAn
Hey, I think this is due to you specified the `norm=batch` in your train_opt.txt. In the code, if you apply the batchNom instead of the instanceNorm, then the bias in...
@kartik1395 Hi, could you please share your scripts to export the TSN model to TensorRT, and how you feed your data to TensorRT runtime to execute?
> > Hi, could you please share your scripts to export the TSN model to TensorRT > > PyTorch -> ONNX -> TensorRT > Convert ONNX to TensorRT with `trtexec`...
Ok, got it. Come to this repo recently, please forgive me if bothering.
The reason why you meet this issue is you are using Windows OS. Just find out where we defined the file path, and modify the “train-images-idx3-ubyte” to “train-images.idx3-ubyte”
Come across this problem also. 
I found many `ptxas` outputs when building `faiss` ``` [ 69%] Building CUDA object faiss/CMakeFiles/faiss.dir/gpu/impl/IVFUtilsSelect1.cu.o ptxas /tmp/tmpxft_00007eae_00000000-6_IVFPQ.ptx, line 4118; warning : ld ptxas /tmp/tmpxft_00007eae_00000000-6_IVFPQ.ptx, line 4278; warning : ld ptxas...
I've tried the CUDA 11.2, but result is pretty much the same. BTW, on your 3090 machine, did you build as we discussed in (#95)? I will really appreciate it...
I didn't install from conda binaries, but from source instead. But the performance on RTX3090 was still not good. Will try to use the conda source later, I am a...
I used Set91 and BSD200 as training set. Traning data 200 epochs on Colab nearly cost 5 hrs