remove C++ deployment example that runs caffe2 format

Summary: n/a

Reviewed By: zhanghang1989

Differential Revision: D32602257

fbshipit-source-id: 44f72c807a8150b2fa50058eb5ad7aad673cd6d9

docker/deploy.Dockerfile

@@ -12,14 +12,7 @@ WORKDIR $HOME
 # Let torchvision find libtorch
 ENV CMAKE_PREFIX_PATH=$HOME/.local/lib/python3.6/site-packages/torch/
 
-RUN sudo apt-get update && sudo apt-get install libgflags-dev libgoogle-glog-dev libopencv-dev --yes
-RUN pip install mkl-include
-
-# Install the correct version of protobuf (find it at torch/caffe2/proto/caffe2.pb.h after installing pytorch):
-RUN wget https://github.com/protocolbuffers/protobuf/releases/download/v3.13.0/protobuf-cpp-3.13.0.tar.gz && tar xf protobuf-cpp-3.13.0.tar.gz
-RUN export CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=$(python3 -c 'import torch; print(int(torch.compiled_with_cxx11_abi()))'); \
-	cd protobuf-3.13.0 && \
-	./configure --prefix=$HOME/.local && make -j && make install
+RUN sudo apt-get update && sudo apt-get install libopencv-dev --yes
 
 # install libtorchvision
 RUN git clone --branch v0.11.1 https://github.com/pytorch/vision/

tools/deploy/CMakeLists.txt

@@ -1,20 +1,12 @@
 # Copyright (c) Facebook, Inc. and its affiliates.
 # See https://pytorch.org/tutorials/advanced/cpp_frontend.html
 cmake_minimum_required(VERSION 3.12 FATAL_ERROR)
-project(caffe2_mask_rcnn)
+project(torchscript_mask_rcnn)
 
 find_package(Torch REQUIRED)
-find_package(gflags REQUIRED)  # needed by caffe2
 find_package(OpenCV REQUIRED)
 find_package(TorchVision REQUIRED)   # needed by export-method=tracing/scripting
 
-add_executable(caffe2_mask_rcnn caffe2_mask_rcnn.cpp)
-target_link_libraries(
-  caffe2_mask_rcnn
-  "${TORCH_LIBRARIES}" gflags glog protobuf ${OpenCV_LIBS})
-set_property(TARGET caffe2_mask_rcnn PROPERTY CXX_STANDARD 14)
-
-
 add_executable(torchscript_mask_rcnn torchscript_mask_rcnn.cpp)
 target_link_libraries(
   torchscript_mask_rcnn

tools/deploy/README.md

@@ -3,27 +3,21 @@ for some high-level background about deployment.
 
 This directory contains the following examples:
 
-1. An example script `export_model.py` (previously called `caffe2_converter.py`)
+1. An example script `export_model.py`
    that exports a detectron2 model for deployment using different methods and formats.
 
-2. A few C++ examples that run inference with Mask R-CNN model in Caffe2/TorchScript format.
+2. A C++ example that runs inference with Mask R-CNN model in TorchScript format.
 
 ## Build
-All C++ examples depend on libtorch and OpenCV. Some require more dependencies:
-
-* Running caffe2-format models requires:
-  * libtorch built with caffe2 inside
-  * gflags, glog
-  * protobuf library that matches the version used by PyTorch (version defined in `include/caffe2/proto/caffe2.pb.h` of your PyTorch installation)
-  * MKL headers if caffe2 is built with MKL
-* Running TorchScript-format models produced by `--export-method=caffe2_tracing` requires no other dependencies.
-* Running TorchScript-format models produced by `--export-method=tracing` requires libtorchvision (C++ library of torchvision).
-
-We build all examples with one `CMakeLists.txt` that requires all the above dependencies.
-Adjust it if you only need one example.
-As a reference,
-we provide a [Dockerfile](../../docker/deploy.Dockerfile) that
-installs all the above dependencies and builds the C++ examples.
+Deployment depends on libtorch and OpenCV. Some require more dependencies:
+
+* Running TorchScript-format models produced by `--export-method=caffe2_tracing` requires libtorch
+  to be built with caffe2 enabled.
+* Running TorchScript-format models produced by `--export-method=tracing/scripting` requires libtorchvision (C++ library of torchvision).
+
+All methods are supported in one C++ file that requires all the above dependencies.
+Adjust it and remove code you don't need.
+As a reference, we provide a [Dockerfile](../../docker/deploy.Dockerfile) that installs all the above dependencies and builds the C++ example.
 
 ## Use
 
@@ -59,26 +53,14 @@ We show a few example commands to export and execute a Mask R-CNN model in C++.
 ```
 
 
-* `export-method=caffe2_tracing, format=caffe2` (caffe2 format will be deprecated):
-```
-./export_model.py --config-file ../../configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml \
-    --output ./output --export-method caffe2_tracing --format caffe2 \
-    MODEL.WEIGHTS detectron2://COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl \
-    MODEL.DEVICE cpu
-
-./build/caffe2_mask_rcnn --predict_net=output/model.pb --init_net=output/model_init.pb --input=input.jpg
-```
-
-
 ## Notes:
 
 1. Tracing/Caffe2-tracing requires valid weights & sample inputs.
    Therefore the above commands require pre-trained models and [COCO dataset](https://detectron2.readthedocs.io/tutorials/builtin_datasets.html).
    You can modify the script to obtain sample inputs in other ways instead of from COCO.
 
-2. `--run-eval` is implemented only for certain modes
-   (caffe2_tracing with caffe2 format, or tracing with torchscript format)
+2. `--run-eval` is implemented only for tracing mode
    to evaluate the exported model using the dataset in the config.
    It's recommended to always verify the accuracy in case the conversion is not successful.
    Evaluation can be slow if model is exported to CPU or dataset is too large ("coco_2017_val_100" is a small subset of COCO useful for evaluation).
-   Caffe2 accuracy may be slightly different (within 0.1 AP) from original model due to numerical precisions between different runtime.
+   `caffe2_tracing` accuracy may be slightly different (within 0.1 AP) from original model due to numerical precisions between different runtime.

tools/deploy/export_model.py

@@ -177,13 +177,13 @@ def get_sample_inputs(args):
         "--format",
         choices=["caffe2", "onnx", "torchscript"],
         help="output format",
-        default="caffe2",
+        default="torchscript",
     )
     parser.add_argument(
         "--export-method",
         choices=["caffe2_tracing", "tracing", "scripting"],
         help="Method to export models",
-        default="caffe2_tracing",
+        default="tracing",
     )
     parser.add_argument("--config-file", default="", metavar="FILE", help="path to config file")
     parser.add_argument("--sample-image", default=None, type=str, help="sample image for input")