ModulatedDeformConv is not implemented on CPU despite being within GPU environment

I am running into this issue when trying to do inference with this simple snippet

image = cv2.imread(image_path, cv2.IMREAD_UNCHANGED).astype(np.float32)
image = np.transpose(image, (-1, 0, 1))  # channels first
norm = transforms.Normalize([123.675, 116.28, 103.53], [58.395, 57.12, 57.375])
image = norm(torch.from_numpy(image))
image = torch.unsqueeze(image, axis=0)  # batch axis

res = vfnet.forward([image], [[{}]], return_loss=False)
print(res)

Which returns RuntimeError: ModulatedDeformConv is not implemented on CPU

This code is run within docker in which mmcv-full with installed this way : pip install mmcv-full==1.2.7 -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html

And mmdetection was installed the same way it is done in the current version of mmdetection's Dockefile.

The result of collect_env is the following:


python mmdet/utils/collect_env.py
sys.platform: linux
Python: 3.7.7 (default, May  7 2020, 21:25:33) [GCC 7.3.0]
CUDA available: True
GPU 0: GeForce GTX 1070 with Max-Q Design
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 10.1, V10.1.243
GCC: gcc (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0
PyTorch: 1.6.0
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.1 Product Build 20200208 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: 4.5.3
MMCV: 1.2.7
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 10.1
MMDetection: 2.10.0+bc521d4

I was able to train this exact same network using VarifocalNet/tools/dist_train.sh. I can send the configuration used to train the model if necessary.

I do not figure out why it is still trying to run on CPU. May you help please ?

open-mmlab / mmdetection

ModulatedDeformConv is not implemented on CPU despite being within GPU environment #5711