Installing caffe on jetson xavier NX

I am using Jetson Xavier NX board having jetpack version 4.5.1-b17

I have installed caffe using sudo apt install caffe-cpu command and when i run which command it shows me /usr/bin/caffe as installation directory.

I am trying to run mnist data set using caffe and I have downloaded its master folder. When I try to run these commands i am getting error.

cd $CAFFE_ROOT
./data/mnist/get_mnist.sh
./examples/mnist/create_mnist.sh

Creating lmdb… ./create_mnist.sh: 17: ./create_mnist.sh: build/examples/mnist/convert_mnist_data.bin: not found

I have read some forums and they say that first set cafferoot directory but i am not sure which is the root directory in my case. I am having some trouble is setting root directory if I am to work with the mnist dataset.

Any guide line on how I can correct this issue ?

Hi @arsalanmajid116,

This topic sounds like it should be placed into the Jetson Xavier NX forum. I will move it over for you.

Tom K

Hi,

It’s more recommended to compile the Caffe frameworks from the source.
Below is a tutorial from the community for your reference:

https://jkjung-avt.github.io/caffe-on-tx2/

Thanks.

I noticed that jet pack 4.6 has support for open CV 4.1.1 but read some where that caffe do not have support for opencv 4.x.x version yet. Should I revert back to some older Cv version 3.x.x or something ?

Hi, I wanted to know if the caffe for jetsontx2 be installed on jetson nx? Does it work

Thanks

Hi, both

Please check the following steps that installing caffe on JetPack 4.6.
(compatible with OpenCV 4.x, cuDNN 8.x and python 3.x)

1. Dependencies

$ sudo apt-get install libboost-dev libboost-all-dev
$ sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev libatlas-base-dev liblmdb-dev libblas-dev libatlas-base-dev libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler

2. Update source

$ git clone https://github.com/BVLC/caffe
$ cd caffe

Apply following patch

diff --git a/Makefile b/Makefile
index b7660e85..cd846e31 100644
--- a/Makefile
+++ b/Makefile
@@ -205,7 +205,7 @@ ifeq ($(USE_OPENCV), 1)
 	endif
 
 endif
-PYTHON_LIBRARIES ?= boost_python python2.7
+PYTHON_LIBRARIES ?= boost_python3 python3.6m
 WARNINGS := -Wall -Wno-sign-compare
 
 ##############################
diff --git a/Makefile.config.example b/Makefile.config.example
index 24ca6327..153a5e60 100644
--- a/Makefile.config.example
+++ b/Makefile.config.example
@@ -2,7 +2,7 @@
 # Contributions simplifying and improving our build system are welcome!
 
 # cuDNN acceleration switch (uncomment to build with cuDNN).
-# USE_CUDNN := 1
+USE_CUDNN := 1
 
 # CPU-only switch (uncomment to build without GPU support).
 # CPU_ONLY := 1
@@ -20,7 +20,7 @@
 # ALLOW_LMDB_NOLOCK := 1
 
 # Uncomment if you're using OpenCV 3
-# OPENCV_VERSION := 3
+OPENCV_VERSION := 3
 
 # To customize your choice of compiler, uncomment and set the following.
 # N.B. the default for Linux is g++ and the default for OSX is clang++
@@ -36,15 +36,8 @@ CUDA_DIR := /usr/local/cuda
 # For CUDA < 6.0, comment the *_50 through *_61 lines for compatibility.
 # For CUDA < 8.0, comment the *_60 and *_61 lines for compatibility.
 # For CUDA >= 9.0, comment the *_20 and *_21 lines for compatibility.
-CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \
-		-gencode arch=compute_20,code=sm_21 \
-		-gencode arch=compute_30,code=sm_30 \
-		-gencode arch=compute_35,code=sm_35 \
-		-gencode arch=compute_50,code=sm_50 \
-		-gencode arch=compute_52,code=sm_52 \
-		-gencode arch=compute_60,code=sm_60 \
-		-gencode arch=compute_61,code=sm_61 \
-		-gencode arch=compute_61,code=compute_61
+CUDA_ARCH := -gencode arch=compute_72,code=sm_72 \
+		-gencode arch=compute_72,code=compute_72
 
 # BLAS choice:
 # atlas for ATLAS (default)
@@ -68,8 +61,8 @@ BLAS := atlas
 
 # NOTE: this is required only if you will compile the python interface.
 # We need to be able to find Python.h and numpy/arrayobject.h.
-PYTHON_INCLUDE := /usr/include/python2.7 \
-		/usr/lib/python2.7/dist-packages/numpy/core/include
+PYTHON_INCLUDE := /usr/include/python3.6 \
+		/usr/lib/python3.6/dist-packages/numpy/core/include
 # Anaconda Python distribution is quite popular. Include path:
 # Verify anaconda location, sometimes it's in root.
 # ANACONDA_HOME := $(HOME)/anaconda
@@ -78,9 +71,9 @@ PYTHON_INCLUDE := /usr/include/python2.7 \
 		# $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include
 
 # Uncomment to use Python 3 (default is Python 2)
-# PYTHON_LIBRARIES := boost_python3 python3.5m
-# PYTHON_INCLUDE := /usr/include/python3.5m \
-#                 /usr/lib/python3.5/dist-packages/numpy/core/include
+PYTHON_LIBRARIES := boost_python3 python3.6m
+PYTHON_INCLUDE := /usr/include/python3.6m \
+                 /usr/lib/python3.6/dist-packages/numpy/core/include
 
 # We need to be able to find libpythonX.X.so or .dylib.
 PYTHON_LIB := /usr/lib
@@ -94,8 +87,8 @@ PYTHON_LIB := /usr/lib
 # WITH_PYTHON_LAYER := 1
 
 # Whatever else you find you need goes here.
-INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include
-LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib
+INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial /usr/include/opencv4
+LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/aarch64-linux-gnu /usr/lib/aarch64-linux-gnu/hdf5/serial
 
 # If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies
 # INCLUDE_DIRS += $(shell brew --prefix)/include
diff --git a/src/caffe/layers/cudnn_conv_layer.cpp b/src/caffe/layers/cudnn_conv_layer.cpp
index efc9e04e..6241f763 100644
--- a/src/caffe/layers/cudnn_conv_layer.cpp
+++ b/src/caffe/layers/cudnn_conv_layer.cpp
@@ -127,45 +127,77 @@ void CuDNNConvolutionLayer<Dtype>::Reshape(
         filter_desc_, pad_h, pad_w,
         stride_h, stride_w);
 
-    // choose forward and backward algorithms + workspace(s)
+#if CUDNN_MAJOR >= 8
+    int returnedAlgoCount;
+    cudnnConvolutionFwdAlgoPerf_t       fw_results[2 * CUDNN_CONVOLUTION_FWD_ALGO_COUNT];
+    cudnnConvolutionBwdDataAlgoPerf_t   bd_results[2 * CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT];
+    cudnnConvolutionBwdFilterAlgoPerf_t bf_results[2 * CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT];
+
+    CUDNN_CHECK(cudnnFindConvolutionForwardAlgorithm(handle_[0],
+        bottom_descs_[i],
+        filter_desc_,
+        conv_descs_[i],
+        top_descs_[i],
+        CUDNN_CONVOLUTION_FWD_ALGO_COUNT,
+        &returnedAlgoCount,
+        fw_results));
+    fwd_algo_[i] = fw_results[0].algo;
+
+    CUDNN_CHECK(cudnnFindConvolutionBackwardDataAlgorithm(handle_[0],
+        filter_desc_,
+        top_descs_[i],
+        conv_descs_[i],
+        bottom_descs_[i],
+        CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT,
+        &returnedAlgoCount,
+        bd_results));
+    bwd_data_algo_[i] = bd_results[0].algo;
+
+    CUDNN_CHECK(cudnnFindConvolutionBackwardFilterAlgorithm(handle_[0],
+        bottom_descs_[i],
+        top_descs_[i],
+        conv_descs_[i],
+        filter_desc_,
+        CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT,
+        &returnedAlgoCount,
+        bf_results));
+    bwd_filter_algo_[i] = bf_results[0].algo;
+#else
     CUDNN_CHECK(cudnnGetConvolutionForwardAlgorithm(handle_[0],
-      bottom_descs_[i],
-      filter_desc_,
-      conv_descs_[i],
-      top_descs_[i],
-      CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
-      workspace_limit_bytes,
-      &fwd_algo_[i]));
+        bottom_descs_[i],
+        filter_desc_,
+        conv_descs_[i],
+        top_descs_[i],
+        CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
+        workspace_limit_bytes,
+        &fwd_algo_[i]));
 
-    CUDNN_CHECK(cudnnGetConvolutionForwardWorkspaceSize(handle_[0],
-      bottom_descs_[i],
-      filter_desc_,
-      conv_descs_[i],
-      top_descs_[i],
-      fwd_algo_[i],
-      &(workspace_fwd_sizes_[i])));
-
-    // choose backward algorithm for filter
     CUDNN_CHECK(cudnnGetConvolutionBackwardFilterAlgorithm(handle_[0],
-          bottom_descs_[i], top_descs_[i], conv_descs_[i], filter_desc_,
-          CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
-          workspace_limit_bytes, &bwd_filter_algo_[i]) );
-
-    // get workspace for backwards filter algorithm
-    CUDNN_CHECK(cudnnGetConvolutionBackwardFilterWorkspaceSize(handle_[0],
-          bottom_descs_[i], top_descs_[i], conv_descs_[i], filter_desc_,
-          bwd_filter_algo_[i], &workspace_bwd_filter_sizes_[i]));
+        bottom_descs_[i], top_descs_[i], conv_descs_[i], filter_desc_,
+        CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
+        workspace_limit_bytes, &bwd_filter_algo_[i]) );
 
-    // choose backward algo for data
     CUDNN_CHECK(cudnnGetConvolutionBackwardDataAlgorithm(handle_[0],
-          filter_desc_, top_descs_[i], conv_descs_[i], bottom_descs_[i],
-          CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT,
+        filter_desc_, top_descs_[i], conv_descs_[i], bottom_descs_[i],
+        CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT,
         workspace_limit_bytes, &bwd_data_algo_[i]));
+#endif
+
+    CUDNN_CHECK(cudnnGetConvolutionForwardWorkspaceSize(handle_[0],
+        bottom_descs_[i],
+        filter_desc_,
+        conv_descs_[i],
+        top_descs_[i],
+        fwd_algo_[i],
+        &(workspace_fwd_sizes_[i])));
+
+    CUDNN_CHECK(cudnnGetConvolutionBackwardFilterWorkspaceSize(handle_[0],
+        bottom_descs_[i], top_descs_[i], conv_descs_[i], filter_desc_,
+        bwd_filter_algo_[i], &workspace_bwd_filter_sizes_[i]));
 
-    // get workspace size
     CUDNN_CHECK(cudnnGetConvolutionBackwardDataWorkspaceSize(handle_[0],
-          filter_desc_, top_descs_[i], conv_descs_[i], bottom_descs_[i],
-          bwd_data_algo_[i], &workspace_bwd_data_sizes_[i]) );
+        filter_desc_, top_descs_[i], conv_descs_[i], bottom_descs_[i],
+        bwd_data_algo_[i], &workspace_bwd_data_sizes_[i]) );
   }
 
   // reduce over all workspace sizes to get a maximum to allocate / reallocate
diff --git a/src/caffe/layers/cudnn_deconv_layer.cpp b/src/caffe/layers/cudnn_deconv_layer.cpp
index 260da5c1..0875ab25 100644
--- a/src/caffe/layers/cudnn_deconv_layer.cpp
+++ b/src/caffe/layers/cudnn_deconv_layer.cpp
@@ -141,7 +141,21 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
                                      stride_h,
                                      stride_w);
 
-    // choose forward and backward algorithms + workspace(s)
+#if CUDNN_MAJOR >= 8
+    int returnedAlgoCount;
+    cudnnConvolutionFwdAlgoPerf_t       fw_results[2 * CUDNN_CONVOLUTION_FWD_ALGO_COUNT];
+
+    CUDNN_CHECK(cudnnFindConvolutionForwardAlgorithm(
+        handle_[0],
+        top_descs_[i],
+        filter_desc_,
+        conv_descs_[i],
+        bottom_descs_[i],
+        CUDNN_CONVOLUTION_FWD_ALGO_COUNT,
+        &returnedAlgoCount,
+        fw_results));
+    fwd_algo_[i] = fw_results[0].algo;
+#else
     CUDNN_CHECK(cudnnGetConvolutionForwardAlgorithm(
         handle_[0],
         top_descs_[i],
@@ -151,6 +165,7 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
         CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
         workspace_limit_bytes,
         &fwd_algo_[i]));
+#endif
 
     // We have found that CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_PRECOMP_GEMM is
     // buggy. Thus, if this algo was chosen, choose winograd instead. If
@@ -183,6 +198,19 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
         fwd_algo_[i],
         &(workspace_fwd_sizes_[i])));
 
+#if CUDNN_MAJOR >= 8
+    cudnnConvolutionBwdFilterAlgoPerf_t bf_results[2 * CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT];
+    CUDNN_CHECK(cudnnFindConvolutionBackwardFilterAlgorithm(
+        handle_[0],
+        top_descs_[i],
+        bottom_descs_[i],
+        conv_descs_[i],
+        filter_desc_,
+        CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT,
+        &returnedAlgoCount,
+        bf_results));
+    bwd_filter_algo_[i] = bf_results[0].algo;
+#else
     // choose backward algorithm for filter
     CUDNN_CHECK(cudnnGetConvolutionBackwardFilterAlgorithm(
         handle_[0],
@@ -193,6 +221,7 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
         CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
         workspace_limit_bytes,
         &bwd_filter_algo_[i]));
+#endif
 
     // get workspace for backwards filter algorithm
     CUDNN_CHECK(cudnnGetConvolutionBackwardFilterWorkspaceSize(
@@ -203,7 +232,19 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
         filter_desc_,
         bwd_filter_algo_[i],
         &workspace_bwd_filter_sizes_[i]));
-
+#if CUDNN_MAJOR >= 8
+    cudnnConvolutionBwdDataAlgoPerf_t   bd_results[2 * CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT];
+    CUDNN_CHECK(cudnnFindConvolutionBackwardDataAlgorithm(
+        handle_[0],
+        filter_desc_,
+        bottom_descs_[i],
+        conv_descs_[i],
+        top_descs_[i],
+        CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT,
+        &returnedAlgoCount,
+        bd_results));
+    bwd_data_algo_[i] = bd_results[0].algo;
+#else
     // choose backward algo for data
     CUDNN_CHECK(cudnnGetConvolutionBackwardDataAlgorithm(
         handle_[0],
@@ -214,6 +255,7 @@ void CuDNNDeconvolutionLayer<Dtype>::Reshape(
         CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT,
         workspace_limit_bytes,
         &bwd_data_algo_[i]));
+#endif
 
     // get workspace size
     CUDNN_CHECK(cudnnGetConvolutionBackwardDataWorkspaceSize(
diff --git a/src/caffe/layers/window_data_layer.cpp b/src/caffe/layers/window_data_layer.cpp
index 1bf3760e..f41169de 100644
--- a/src/caffe/layers/window_data_layer.cpp
+++ b/src/caffe/layers/window_data_layer.cpp
@@ -290,7 +290,7 @@ void WindowDataLayer<Dtype>::load_batch(Batch<Dtype>* batch) {
           image_database_cache_[window[WindowDataLayer<Dtype>::IMAGE_INDEX]];
         cv_img = DecodeDatumToCVMat(image_cached.second, true);
       } else {
-        cv_img = cv::imread(image.first, CV_LOAD_IMAGE_COLOR);
+        cv_img = cv::imread(image.first, cv::IMREAD_COLOR);
         if (!cv_img.data) {
           LOG(ERROR) << "Could not open or find file " << image.first;
           return;
diff --git a/src/caffe/util/io.cpp b/src/caffe/util/io.cpp
index 5295d9dd..ca8a6492 100644
--- a/src/caffe/util/io.cpp
+++ b/src/caffe/util/io.cpp
@@ -73,8 +73,8 @@ void WriteProtoToBinaryFile(const Message& proto, const char* filename) {
 cv::Mat ReadImageToCVMat(const string& filename,
     const int height, const int width, const bool is_color) {
   cv::Mat cv_img;
-  int cv_read_flag = (is_color ? CV_LOAD_IMAGE_COLOR :
-    CV_LOAD_IMAGE_GRAYSCALE);
+  int cv_read_flag = (is_color ? cv::IMREAD_COLOR :
+    cv::ImreadModes::IMREAD_GRAYSCALE);
   cv::Mat cv_img_origin = cv::imread(filename, cv_read_flag);
   if (!cv_img_origin.data) {
     LOG(ERROR) << "Could not open or find file " << filename;
@@ -179,8 +179,8 @@ cv::Mat DecodeDatumToCVMat(const Datum& datum, bool is_color) {
   CHECK(datum.encoded()) << "Datum not encoded";
   const string& data = datum.data();
   std::vector<char> vec_data(data.c_str(), data.c_str() + data.size());
-  int cv_read_flag = (is_color ? CV_LOAD_IMAGE_COLOR :
-    CV_LOAD_IMAGE_GRAYSCALE);
+  int cv_read_flag = (is_color ? cv::IMREAD_COLOR :
+    cv::ImreadModes::IMREAD_GRAYSCALE);
   cv_img = cv::imdecode(vec_data, cv_read_flag);
   if (!cv_img.data) {
     LOG(ERROR) << "Could not decode datum ";
$ cp Makefile.config.example Makefile.config
$ make -j4

3. Setup for python3

$ cd src/
$ wget https://pypi.python.org/packages/03/98/1521e7274cfbcc678e9640e242a62cbcd18743f9c5761179da165c940eac/leveldb-0.20.tar.gz
$ tar xzvf leveldb-0.20.tar.gz
$ cd leveldb-0.20
$ python3 setup.py build
$ sudo apt-get install gfortran
$ pkgs=`sed 's/[>=<].*$//' ../../python/requirements.txt`
$ for pkg in $pkgs; do sudo pip3 install $pkg; done
$ cd ../../
$ make pycaffe

4. Test

$ export PYTHONPATH=${PWD}/python
$ python3
Python 3.6.9 (default, Jan 26 2021, 15:33:00)
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import caffe
>>> caffe.__version__
'1.0.0'

Refer to How to Install Caffe and PyCaffe on Jetson TX2
Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.