Random seed matters?

[VingtDylan]

I cannot get the desired results(epe = 0.78 on flyingchairs), I check the log file provided and find that the seed is "null", so what does the null value means? Doesn't the function "init_random_seed" generate a seed?

[VingtDylan]

my configuration shows below:

2021-11-25 03:12:02,022 - mmflow - INFO - Environment info:
------------------------------------------------------------
sys.platform: linux
Python: 3.7.11 (default, Jul 27 2021, 14:32:16) [GCC 7.5.0]
CUDA available: True
CUDA_HOME: :/usr/local/cuda-11.0:/usr/local/cuda-11.0:/usr/local/cuda-11.0
GPU 0,1,2,3: GeForce RTX 2080 Ti
GCC: gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
PyTorch: 1.6.0
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v1.5.0 (Git Hash e2ac1fac44c5078ca927cb9b90e1b3066a0b2ed0)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 10.1
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_37,code=compute_37
  - CuDNN 7.6.3
  - Magma 2.5.2
  - Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DUSE_VULKAN_WRAPPER -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_STATIC_DISPATCH=OFF, 

TorchVision: 0.7.0
OpenCV: 4.5.4
MMCV: 1.3.17
MMOpticalFlow: 0.1.0+c7ebc17
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 10.1
------------------------------------------------------------

I follow the guidance and then test some seeds, only to find that EPE on flyingchairs is around 1.01.

I execute the command such as:

CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29500 ./tools/dist_train.sh configs/raft/raft_8x2_100k_flyingchairs_368x496.py 4 --work-dir output/raft/G4-xxx

I am looking forward to your guidance.

[MeowZheng]

I saw you used 4 GPUs when training, but we used 8 GPUs. raft_8x2_100k_flyingchairs_368x496 means training RAFT model on FlyingChairs with size (368, 496), and using 8 GPUs and 2 batch/GPUs, so the batchsize is 16, not 8 from your setting.

We didn't use init_random_seed to generate a seed when training RAFT. You can see we added this function in #27 after releasing models, and it works for making all ranks have the same seed if users didn't provide seed.

[VingtDylan]

Thanks a lot.

[MeowZheng]

Glad to help