Merge pull request #797 from exadel-inc/cuda-11.x

Cuda 11.x

embedding-calculator/Dockerfile

@@ -3,7 +3,7 @@ FROM ${BASE_IMAGE:-python:3.7-slim}
 
 RUN apt-get update && apt-get install -y build-essential cmake git wget unzip \
         curl yasm pkg-config libswscale-dev libtbb2 libtbb-dev libjpeg-dev \
-        libpng-dev libtiff-dev libavformat-dev libpq-dev libfreeimage3 \
+        libpng-dev libtiff-dev libavformat-dev libpq-dev libfreeimage3 python3-opencv \
     && rm -rf /var/lib/apt/lists/*
 
 # install common python packages
@@ -49,4 +49,4 @@ EXPOSE 3000
 COPY uwsgi.ini .
 ENV UWSGI_PROCESSES=${UWSGI_PROCESSES:-2}
 ENV UWSGI_THREADS=1
-CMD ["uwsgi", "--ini", "uwsgi.ini"]
+CMD ["uwsgi", "--ini", "uwsgi.ini"]

embedding-calculator/Makefile

@@ -2,8 +2,9 @@ SHELL := /bin/bash
 .DEFAULT_GOAL := default
 .PHONY := default up build-images build-cuda
 
+VERSION := 1.0.1
 IMAGE := ${DOCKER_REGISTRY}compreface-core
-CUDA_IMAGE = $(IMAGE)-base:base-cuda100-py37
+CUDA_IMAGE = $(IMAGE)-base:base-cuda112-py37
 APPERY_ARG := --build-arg APPERY_API_KEY=${APPERY_API_KEY}
 
 MOBILENET_BUILD_ARGS := --build-arg FACE_DETECTION_PLUGIN=insightface.FaceDetector@retinaface_mnet025_v1 \
@@ -70,3 +71,4 @@ tools/tmp:
 tools/benchmark_detection/tmp:
 	$(call get_from_remote_tgz,http://tamaraberg.com/faceDataset/originalPics.tar.gz,tools/benchmark_detection/tmp/originalPics)
 	$(call get_from_remote_tgz,http://vis-www.cs.umass.edu/fddb/FDDB-folds.tgz,tools/benchmark_detection/tmp)
+

embedding-calculator/README.md

@@ -162,7 +162,14 @@ There are two build arguments for optimization:
 * `INTEL_OPTIMIZATION` - enable Intel MKL optimization (true/false)
 
 
-##### NVIDIA Runtime   
+##### GPU Setup (Windows):
+1. Install or update Docker Desktop.
+2. Make sure that you have Windows version 21H2 or higher.
+3. Update your NVIDIA drivers.
+4. Install or update WSL2 Linux kernel.
+5. Make sure the WSL2 backend is enabled in Docker Desktop.
+
+##### GPU Setup (Linux): 
 
 Install the nvidia-docker2 package and dependencies on the host machine:
 ```
@@ -171,7 +178,7 @@ sudo apt-get install -y nvidia-docker2
 sudo systemctl restart docker
 ```
 
-Build and run with enabled gpu
+##### Build and run with enabled gpu
 ```
 docker build . -t embedding-calculator-cuda -f gpu.Dockerfile
 docker build . -t embedding-calculator-gpu --build-arg GPU_IDX=0 --build-arg BASE_IMAGE=embedding-calculator-cuda

embedding-calculator/gpu.Dockerfile

@@ -1,70 +1,27 @@
-ARG UBUNTU_VERSION=18.04
+ARG BASE_IMAGE
+FROM ${BASE_IMAGE:-nvidia/cuda:11.2.2-cudnn8-runtime-ubuntu20.04}
 
-ARG ARCH=
-ARG CUDA=10.0
-FROM nvidia/cuda${ARCH:+-$ARCH}:${CUDA}-base-ubuntu${UBUNTU_VERSION} as base
-# ARCH and CUDA are specified again because the FROM directive resets ARGs
-# (but their default value is retained if set previously)
-ARG ARCH
-ARG CUDA
-ARG CUDNN=7.6.4.38-1
-ARG CUDNN_MAJOR_VERSION=7
-ARG LIB_DIR_PREFIX=x86_64
-ARG LIBNVINFER=6.0.1-1
-ARG LIBNVINFER_MAJOR_VERSION=6
-ENV CUDA=$CUDA
+ENV DEBIAN_FRONTEND=noninteractive
+ENV CUDA=11.2
 
-# Needed for string substitution
-SHELL ["/bin/bash", "-c"]
-# Pick up some TF dependencies
 RUN apt-get update && apt-get install -y --no-install-recommends \
         build-essential \
-        cuda-command-line-tools-${CUDA/./-} \
-        # There appears to be a regression in libcublas10=10.2.2.89-1 which
-        # prevents cublas from initializing in TF. See
-        # https://github.com/tensorflow/tensorflow/issues/9489#issuecomment-562394257
-        libcublas10=10.2.1.243-1 \
-        cuda-nvrtc-${CUDA/./-} \
-        cuda-cufft-${CUDA/./-} \
-        cuda-curand-${CUDA/./-} \
-        cuda-cusolver-${CUDA/./-} \
-        cuda-cusparse-${CUDA/./-} \
-        curl \
-        libcudnn7=${CUDNN}+cuda${CUDA} \
-        libfreetype6-dev \
-        libhdf5-serial-dev \
-        libzmq3-dev \
-        pkg-config \
         software-properties-common \
-        unzip
-
-# Install TensorRT if not building for PowerPC
-RUN [[ "${ARCH}" = "ppc64le" ]] || { apt-get update && \
-        apt-get install -y --no-install-recommends libnvinfer${LIBNVINFER_MAJOR_VERSION}=${LIBNVINFER}+cuda${CUDA} \
-        libnvinfer-plugin${LIBNVINFER_MAJOR_VERSION}=${LIBNVINFER}+cuda${CUDA} \
-        && apt-get clean \
-        && rm -rf /var/lib/apt/lists/*; }
-
-# For CUDA profiling, TensorFlow requires CUPTI.
-ENV LD_LIBRARY_PATH /usr/local/cuda/extras/CUPTI/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
-
-# Link the libcuda stub to the location where tensorflow is searching for it and reconfigure
-# dynamic linker run-time bindings
-RUN ln -s /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1 \
-    && echo "/usr/local/cuda/lib64/stubs" > /etc/ld.so.conf.d/z-cuda-stubs.conf \
-    && ldconfig
+		curl \
+		pkg-config \
+		unzip \
+    	python3-dev \
+    	python3-distutils \
+    && rm -rf /var/lib/apt/lists/*
 
 # See http://bugs.python.org/issue19846
 ENV LANG C.UTF-8
 
-ARG PYTHON=python3.7
-RUN add-apt-repository ppa:deadsnakes/ppa && apt-get update && apt-get install -y ${PYTHON} libpython3.7-dev libgl1-mesa-glx
-RUN curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && ${PYTHON} get-pip.py
-RUN ${PYTHON} -m pip --no-cache-dir install --upgrade pip setuptools
+RUN curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && python3 get-pip.py
+RUN python3 -m pip --no-cache-dir install --upgrade pip setuptools
 
 # Some TF tools expect a "python" binary
-RUN ln -s $(which $PYTHON) /usr/local/bin/python
-
+RUN ln -s $(which python3) /usr/local/bin/python
 
 # Variables for Tensorflow
 ENV TF_FORCE_GPU_ALLOW_GROWTH=true

embedding-calculator/requirements.txt

@@ -5,7 +5,7 @@ flasgger==0.9.5
 Flask==1.1.2
 itsdangerous==2.0.1 # itsdangerous is a Flask dependency, update it with Flask
 jinja2<3.1.0 # jinja2 is a Flask dependency, update it with Flask
-gdown~=4.3.1
+gdown~=4.3.0
 Werkzeug==1.0.1
 PyYAML==5.4.1
 

embedding-calculator/src/constants.py

@@ -34,6 +34,8 @@ class ENV(Constants):
     GPU_IDX = int(get_env('GPU_IDX', '-1'))
     INTEL_OPTIMIZATION = get_env_bool('INTEL_OPTIMIZATION')
 
+    RUN_MODE = get_env_bool('RUN_MODE', False)
+
 
 LOGGING_LEVEL = logging._nameToLevel[ENV.LOGGING_LEVEL_NAME]
-ENV_MAIN = ENV
+ENV_MAIN = ENV

embedding-calculator/src/init_runtime.py

@@ -19,6 +19,7 @@
 from PIL import ImageFile
 
 from src._logging import init_logging
+from src.constants import ENV
 
 
 def _check_ci_build_args():
@@ -33,4 +34,5 @@ def init_runtime(logging_level):
     assert sys.version_info >= (3, 7)
     ImageFile.LOAD_TRUNCATED_IMAGES = True
     _check_ci_build_args()
-    init_logging(logging_level)
+    ENV.RUN_MODE = True
+    init_logging(logging_level)

embedding-calculator/src/services/facescan/plugins/dependencies.py

@@ -32,7 +32,5 @@ def get_mxnet() -> Tuple[str, ...]:
     mxnet_lib = 'mxnet-'
     if ENV.GPU_IDX > -1 and cuda_version:
         mxnet_lib += f"cu{cuda_version}"
-    if ENV.INTEL_OPTIMIZATION:
-        mxnet_lib += 'mkl'
     mxnet_lib = mxnet_lib.rstrip('-')
-    return (f'{mxnet_lib}<1.7',)
+    return (f'{mxnet_lib}==1.9.0',)

embedding-calculator/src/services/facescan/plugins/facenet/__init__.py

@@ -14,4 +14,4 @@
 
 from src.services.facescan.plugins.dependencies import get_tensorflow
 
-requirements = get_tensorflow() # + ('mtcnn~=0.1.0',)
+requirements = ('protobuf~=3.20.1',) + get_tensorflow() # + ('mtcnn~=0.1.0',)

embedding-calculator/src/services/facescan/plugins/insightface/facemask/facemask.py

@@ -18,14 +18,17 @@
 from cached_property import cached_property
 
 import numpy as np
-import mxnet as mx
-from mxnet.gluon.model_zoo import vision
-from mxnet.gluon.data.vision import transforms
 
 from src.services.dto import plugin_result
 from src.services.imgtools.types import Array3D
 from src.services.facescan.plugins import base
 from src.services.facescan.plugins.insightface.insightface import InsightFaceMixin
+from src.constants import ENV
+
+if ENV.RUN_MODE:
+    import mxnet as mx
+    from mxnet.gluon.model_zoo import vision
+    from mxnet.gluon.data.vision import transforms
 
 
 class MaskDetector(InsightFaceMixin, base.BasePlugin):
@@ -35,11 +38,12 @@ class MaskDetector(InsightFaceMixin, base.BasePlugin):
         ('mobilenet_v2_on_mafa_kaggle123', '1DYUIroNXkuYKQypYtCxQvAItLnrTTt5E'),
         ('resnet18_on_mafa_kaggle123', '1A3fNrvgrJqMw54cWRj47LNFNnFvTjmdj')
     )
-    img_transforms = transforms.Compose([
-        transforms.Resize(224),
-        transforms.ToTensor(),
-        transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
-    ])
+    if ENV.RUN_MODE:
+        img_transforms = transforms.Compose([
+            transforms.Resize(224),
+            transforms.ToTensor(),
+            transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
+        ])
 
     @property
     def input_image_size(self) -> Tuple[int, int]:

embedding-calculator/src/services/facescan/plugins/insightface/helpers.py

@@ -13,11 +13,16 @@
 #  permissions and limitations under the License.
 
 from typing import Tuple
-import mxnet as mx
 import numpy as np
 import cv2
 from skimage import transform as trans
 
+from src.constants import ENV
+
+
+if ENV.RUN_MODE:
+    import mxnet as mx
+
 
 def predict_landmark2d106(model, img,
                           crop_size:  Tuple[int, int],
@@ -66,4 +71,4 @@ def trans_points2d(pts, M):
         new_pt = np.dot(M, new_pt)
         #print('new_pt', new_pt.shape, new_pt)
         new_pts[i] = new_pt[0:2]
-    return new_pts
+    return new_pts

embedding-calculator/src/services/facescan/plugins/insightface/insightface.py

@@ -17,12 +17,7 @@
 from typing import List, Tuple
 import attr
 import numpy as np
-import mxnet as mx
 from cached_property import cached_property
-from insightface.app import FaceAnalysis
-from insightface.model_zoo import (model_store, face_detection,
-                                   face_recognition, face_genderage)
-from insightface.utils import face_align
 
 from src.constants import ENV
 from src.services.dto.bounding_box import BoundingBoxDTO
@@ -37,6 +32,22 @@
 logger = logging.getLogger(__name__)
 
 
+if ENV.RUN_MODE:
+    import mxnet as mx
+
+    from insightface.app import FaceAnalysis
+    from insightface.model_zoo import (model_store, face_detection,
+                                    face_recognition, face_genderage)
+    from insightface.utils import face_align
+
+    class DetectionOnlyFaceAnalysis(FaceAnalysis):
+        rec_model = None
+        ga_model = None
+
+        def __init__(self, file):
+            self.det_model = face_detection.FaceDetector(file, 'net3')
+
+
 class InsightFaceMixin:
     _CTX_ID = ENV.GPU_IDX
     _NMS = 0.4
@@ -48,14 +59,6 @@ def get_model_file(self, ml_model: base.MLModel):
         return model_store.find_params_file(ml_model.path)
 
 
-class DetectionOnlyFaceAnalysis(FaceAnalysis):
-    rec_model = None
-    ga_model = None
-
-    def __init__(self, file):
-        self.det_model = face_detection.FaceDetector(file, 'net3')
-
-
 class FaceDetector(InsightFaceMixin, mixins.FaceDetectorMixin, base.BasePlugin):
     ml_models = (
         ('retinaface_mnet025_v1', '1ggNFFqpe0abWz6V1A82rnxD6fyxB8W2c'),
@@ -179,7 +182,6 @@ class LandmarksDetector(mixins.LandmarksDetectorMixin, base.BasePlugin):
 class Landmarks2d106DTO(plugin_result.LandmarksDTO):
     """ 
     106-points facial landmarks 
-
     Points mark-up - https://github.com/deepinsight/insightface/tree/master/alignment/coordinateReg#visualization
     """
     NOSE_POSITION = 86
@@ -211,4 +213,4 @@ def _landmark_model(self):
         model.bind(for_training=False,
                    data_shapes=[('data', (1, 3, *self.CROP_SIZE))])
         model.set_params(arg_params, aux_params)
-        return model
+        return model