diff --git a/3rdparty/libjpeg-turbo/CMakeLists.txt b/3rdparty/libjpeg-turbo/CMakeLists.txt
index 3c7f29b08e..73e1eee141 100644
--- a/3rdparty/libjpeg-turbo/CMakeLists.txt
+++ b/3rdparty/libjpeg-turbo/CMakeLists.txt
@@ -4,9 +4,9 @@ ocv_warnings_disable(CMAKE_C_FLAGS -Wunused-parameter -Wsign-compare -Wshorten-6
 
 set(VERSION_MAJOR 2)
 set(VERSION_MINOR 1)
-set(VERSION_REVISION 0)
+set(VERSION_REVISION 2)
 set(VERSION ${VERSION_MAJOR}.${VERSION_MINOR}.${VERSION_REVISION})
-set(LIBJPEG_TURBO_VERSION_NUMBER 2001000)
+set(LIBJPEG_TURBO_VERSION_NUMBER 2001002)
 
 string(TIMESTAMP BUILD "opencv-${OPENCV_VERSION}-libjpeg-turbo")
 if(CMAKE_BUILD_TYPE STREQUAL "Debug")
diff --git a/3rdparty/libjpeg-turbo/jconfigint.h.in b/3rdparty/libjpeg-turbo/jconfigint.h.in
index c94bffaa35..a46979df1e 100644
--- a/3rdparty/libjpeg-turbo/jconfigint.h.in
+++ b/3rdparty/libjpeg-turbo/jconfigint.h.in
@@ -40,3 +40,13 @@
 #define HAVE_BITSCANFORWARD
 #endif
 #endif
+
+#if defined(__has_attribute)
+#if __has_attribute(fallthrough)
+#define FALLTHROUGH  __attribute__((fallthrough));
+#else
+#define FALLTHROUGH
+#endif
+#else
+#define FALLTHROUGH
+#endif
diff --git a/3rdparty/libjpeg-turbo/src/jchuff.c b/3rdparty/libjpeg-turbo/src/jchuff.c
index 2bce767ebd..8ff817b151 100644
--- a/3rdparty/libjpeg-turbo/src/jchuff.c
+++ b/3rdparty/libjpeg-turbo/src/jchuff.c
@@ -44,8 +44,9 @@
  * flags (this defines __thumb__).
  */
 
-#if defined(__arm__) || defined(__aarch64__) || defined(_M_ARM) || \
-    defined(_M_ARM64)
+/* NOTE: Both GCC and Clang define __GNUC__ */
+#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \
+    defined(_M_ARM) || defined(_M_ARM64)
 #if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
diff --git a/3rdparty/libjpeg-turbo/src/jcmaster.c b/3rdparty/libjpeg-turbo/src/jcmaster.c
index 998dc40a5c..c2b2600031 100644
--- a/3rdparty/libjpeg-turbo/src/jcmaster.c
+++ b/3rdparty/libjpeg-turbo/src/jcmaster.c
@@ -493,7 +493,7 @@ prepare_for_pass(j_compress_ptr cinfo)
     master->pass_type = output_pass;
     master->pass_number++;
 #endif
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
   case output_pass:
     /* Do a data-output pass. */
     /* We need not repeat per-scan setup if prior optimization pass did it. */
diff --git a/3rdparty/libjpeg-turbo/src/jcphuff.c b/3rdparty/libjpeg-turbo/src/jcphuff.c
index bd14fc27d5..1101987180 100644
--- a/3rdparty/libjpeg-turbo/src/jcphuff.c
+++ b/3rdparty/libjpeg-turbo/src/jcphuff.c
@@ -7,6 +7,7 @@
  * Copyright (C) 2011, 2015, 2018, 2021, D. R. Commander.
  * Copyright (C) 2016, 2018, Matthieu Darbois.
  * Copyright (C) 2020, Arm Limited.
+ * Copyright (C) 2021, Alex Richardson.
  * For conditions of distribution and use, see the accompanying README.ijg
  * file.
  *
@@ -52,8 +53,9 @@
  * flags (this defines __thumb__).
  */
 
-#if defined(__arm__) || defined(__aarch64__) || defined(_M_ARM) || \
-    defined(_M_ARM64)
+/* NOTE: Both GCC and Clang define __GNUC__ */
+#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \
+    defined(_M_ARM) || defined(_M_ARM64)
 #if !defined(__thumb__) || defined(__thumb2__)
 #define USE_CLZ_INTRINSIC
 #endif
@@ -679,7 +681,7 @@ encode_mcu_AC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
       emit_restart(entropy, entropy->next_restart_num);
 
 #ifdef WITH_SIMD
-  cvalue = values = (JCOEF *)PAD((size_t)values_unaligned, 16);
+  cvalue = values = (JCOEF *)PAD((JUINTPTR)values_unaligned, 16);
 #else
   /* Not using SIMD, so alignment is not needed */
   cvalue = values = values_unaligned;
@@ -944,7 +946,7 @@ encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
       emit_restart(entropy, entropy->next_restart_num);
 
 #ifdef WITH_SIMD
-  cabsvalue = absvalues = (JCOEF *)PAD((size_t)absvalues_unaligned, 16);
+  cabsvalue = absvalues = (JCOEF *)PAD((JUINTPTR)absvalues_unaligned, 16);
 #else
   /* Not using SIMD, so alignment is not needed */
   cabsvalue = absvalues = absvalues_unaligned;
diff --git a/3rdparty/libjpeg-turbo/src/jdapimin.c b/3rdparty/libjpeg-turbo/src/jdapimin.c
index 21a41d2e9f..4609b1322f 100644
--- a/3rdparty/libjpeg-turbo/src/jdapimin.c
+++ b/3rdparty/libjpeg-turbo/src/jdapimin.c
@@ -23,6 +23,7 @@
 #include "jinclude.h"
 #include "jpeglib.h"
 #include "jdmaster.h"
+#include "jconfigint.h"
 
 
 /*
@@ -308,7 +309,7 @@ jpeg_consume_input(j_decompress_ptr cinfo)
     /* Initialize application's data source module */
     (*cinfo->src->init_source) (cinfo);
     cinfo->global_state = DSTATE_INHEADER;
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
   case DSTATE_INHEADER:
     retcode = (*cinfo->inputctl->consume_input) (cinfo);
     if (retcode == JPEG_REACHED_SOS) { /* Found SOS, prepare to decompress */
diff --git a/3rdparty/libjpeg-turbo/src/jdhuff.c b/3rdparty/libjpeg-turbo/src/jdhuff.c
index f786c10547..679d221685 100644
--- a/3rdparty/libjpeg-turbo/src/jdhuff.c
+++ b/3rdparty/libjpeg-turbo/src/jdhuff.c
@@ -584,7 +584,7 @@ decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
        * behavior is, to the best of our understanding, innocuous, and it is
        * unclear how to work around it without potentially affecting
        * performance.  Thus, we (hopefully temporarily) suppress UBSan integer
-       * overflow errors for this function.
+       * overflow errors for this function and decode_mcu_fast().
        */
       s += state.last_dc_val[ci];
       state.last_dc_val[ci] = s;
@@ -651,6 +651,12 @@ decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 }
 
 
+#if defined(__has_feature)
+#if __has_feature(undefined_behavior_sanitizer)
+__attribute__((no_sanitize("signed-integer-overflow"),
+               no_sanitize("unsigned-integer-overflow")))
+#endif
+#endif
 LOCAL(boolean)
 decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
@@ -681,6 +687,9 @@ decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 
     if (entropy->dc_needed[blkn]) {
       int ci = cinfo->MCU_membership[blkn];
+      /* Refer to the comment in decode_mcu_slow() regarding the supression of
+       * a UBSan integer overflow error in this line of code.
+       */
       s += state.last_dc_val[ci];
       state.last_dc_val[ci] = s;
       if (block)
diff --git a/3rdparty/libjpeg-turbo/src/jdmainct.c b/3rdparty/libjpeg-turbo/src/jdmainct.c
index 50301d6b50..f466b259f0 100644
--- a/3rdparty/libjpeg-turbo/src/jdmainct.c
+++ b/3rdparty/libjpeg-turbo/src/jdmainct.c
@@ -18,6 +18,7 @@
 
 #include "jinclude.h"
 #include "jdmainct.h"
+#include "jconfigint.h"
 
 
 /*
@@ -360,7 +361,7 @@ process_data_context_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
     main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
     if (*out_row_ctr >= out_rows_avail)
       return;                   /* Postprocessor exactly filled output buf */
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
   case CTX_PREPARE_FOR_IMCU:
     /* Prepare to process first M-1 row groups of this iMCU row */
     main_ptr->rowgroup_ctr = 0;
@@ -371,7 +372,7 @@ process_data_context_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
     if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
       set_bottom_pointers(cinfo);
     main_ptr->context_state = CTX_PROCESS_IMCU;
-    /*FALLTHROUGH*/
+    FALLTHROUGH                 /*FALLTHROUGH*/
   case CTX_PROCESS_IMCU:
     /* Call postprocessor using previously set pointers */
     (*cinfo->post->post_process_data) (cinfo,
diff --git a/3rdparty/libjpeg-turbo/src/jmemmgr.c b/3rdparty/libjpeg-turbo/src/jmemmgr.c
index 508ca7429c..70b8ec0c49 100644
--- a/3rdparty/libjpeg-turbo/src/jmemmgr.c
+++ b/3rdparty/libjpeg-turbo/src/jmemmgr.c
@@ -4,7 +4,7 @@
  * This file was part of the Independent JPEG Group's software:
  * Copyright (C) 1991-1997, Thomas G. Lane.
  * libjpeg-turbo Modifications:
- * Copyright (C) 2016, D. R. Commander.
+ * Copyright (C) 2016, 2021, D. R. Commander.
  * For conditions of distribution and use, see the accompanying README.ijg
  * file.
  *
@@ -1032,7 +1032,7 @@ free_pool(j_common_ptr cinfo, int pool_id)
     large_pool_ptr next_lhdr_ptr = lhdr_ptr->next;
     space_freed = lhdr_ptr->bytes_used +
                   lhdr_ptr->bytes_left +
-                  sizeof(large_pool_hdr);
+                  sizeof(large_pool_hdr) + ALIGN_SIZE - 1;
     jpeg_free_large(cinfo, (void *)lhdr_ptr, space_freed);
     mem->total_space_allocated -= space_freed;
     lhdr_ptr = next_lhdr_ptr;
@@ -1045,7 +1045,7 @@ free_pool(j_common_ptr cinfo, int pool_id)
   while (shdr_ptr != NULL) {
     small_pool_ptr next_shdr_ptr = shdr_ptr->next;
     space_freed = shdr_ptr->bytes_used + shdr_ptr->bytes_left +
-                  sizeof(small_pool_hdr);
+                  sizeof(small_pool_hdr) + ALIGN_SIZE - 1;
     jpeg_free_small(cinfo, (void *)shdr_ptr, space_freed);
     mem->total_space_allocated -= space_freed;
     shdr_ptr = next_shdr_ptr;
diff --git a/3rdparty/libjpeg-turbo/src/jpegint.h b/3rdparty/libjpeg-turbo/src/jpegint.h
index 195fbcb9b6..8c8534793a 100644
--- a/3rdparty/libjpeg-turbo/src/jpegint.h
+++ b/3rdparty/libjpeg-turbo/src/jpegint.h
@@ -5,8 +5,9 @@
  * Copyright (C) 1991-1997, Thomas G. Lane.
  * Modified 1997-2009 by Guido Vollbeding.
  * libjpeg-turbo Modifications:
- * Copyright (C) 2015-2016, 2019, D. R. Commander.
+ * Copyright (C) 2015-2016, 2019, 2021, D. R. Commander.
  * Copyright (C) 2015, Google, Inc.
+ * Copyright (C) 2021, Alex Richardson.
  * For conditions of distribution and use, see the accompanying README.ijg
  * file.
  *
@@ -47,6 +48,18 @@ typedef enum {            /* Operating modes for buffer controllers */
 /* JLONG must hold at least signed 32-bit values. */
 typedef long JLONG;
 
+/* JUINTPTR must hold pointer values. */
+#ifdef __UINTPTR_TYPE__
+/*
+ * __UINTPTR_TYPE__ is GNU-specific and available in GCC 4.6+ and Clang 3.0+.
+ * Fortunately, that is sufficient to support the few architectures for which
+ * sizeof(void *) != sizeof(size_t).  The only other options would require C99
+ * or Clang-specific builtins.
+ */
+typedef __UINTPTR_TYPE__ JUINTPTR;
+#else
+typedef size_t JUINTPTR;
+#endif
 
 /*
  * Left shift macro that handles a negative operand without causing any
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0ca33755ff..4718c383ae 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -106,7 +106,7 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ${ENABLE_PIC})
 
 ocv_cmake_hook(PRE_CMAKE_BOOTSTRAP)
 
-# Bootstap CMake system: setup CMAKE_SYSTEM_NAME and other vars
+# Bootstrap CMake system: setup CMAKE_SYSTEM_NAME and other vars
 if(OPENCV_WORKAROUND_CMAKE_20989)
   set(CMAKE_SYSTEM_PROCESSOR_BACKUP ${CMAKE_SYSTEM_PROCESSOR})
 endif()
diff --git a/doc/js_tutorials/js_assets/js_camshift.html b/doc/js_tutorials/js_assets/js_camshift.html
index 046ab20efd..b2d10751fa 100644
--- a/doc/js_tutorials/js_assets/js_camshift.html
+++ b/doc/js_tutorials/js_assets/js_camshift.html
@@ -77,7 +77,7 @@ cv.normalize(roiHist, roiHist, 0, 255, cv.NORM_MINMAX);
 // delete useless mats.
 roi.delete(); hsvRoi.delete(); mask.delete(); low.delete(); high.delete(); hsvRoiVec.delete();
 
-// Setup the termination criteria, either 10 iteration or move by atleast 1 pt
+// Setup the termination criteria, either 10 iteration or move by at least 1 pt
 let termCrit = new cv.TermCriteria(cv.TERM_CRITERIA_EPS | cv.TERM_CRITERIA_COUNT, 10, 1);
 
 let hsv = new cv.Mat(video.height, video.width, cv.CV_8UC3);
diff --git a/doc/js_tutorials/js_assets/js_meanshift.html b/doc/js_tutorials/js_assets/js_meanshift.html
index 9e29002c8d..958539bede 100644
--- a/doc/js_tutorials/js_assets/js_meanshift.html
+++ b/doc/js_tutorials/js_assets/js_meanshift.html
@@ -77,7 +77,7 @@ cv.normalize(roiHist, roiHist, 0, 255, cv.NORM_MINMAX);
 // delete useless mats.
 roi.delete(); hsvRoi.delete(); mask.delete(); low.delete(); high.delete(); hsvRoiVec.delete();
 
-// Setup the termination criteria, either 10 iteration or move by atleast 1 pt
+// Setup the termination criteria, either 10 iteration or move by at least 1 pt
 let termCrit = new cv.TermCriteria(cv.TERM_CRITERIA_EPS | cv.TERM_CRITERIA_COUNT, 10, 1);
 
 let hsv = new cv.Mat(video.height, video.width, cv.CV_8UC3);
diff --git a/modules/core/src/logger.cpp b/modules/core/src/logger.cpp
index dbb25f9e09..f395571e80 100644
--- a/modules/core/src/logger.cpp
+++ b/modules/core/src/logger.cpp
@@ -181,17 +181,33 @@ LogLevel getLogLevel()
 
 namespace internal {
 
+static int getShowTimestampMode()
+{
+    static bool param_timestamp_enable = utils::getConfigurationParameterBool("OPENCV_LOG_TIMESTAMP", true);
+    static bool param_timestamp_ns_enable = utils::getConfigurationParameterBool("OPENCV_LOG_TIMESTAMP_NS", false);
+    return (param_timestamp_enable ? 1 : 0) + (param_timestamp_ns_enable ? 2 : 0);
+}
+
 void writeLogMessage(LogLevel logLevel, const char* message)
 {
     const int threadID = cv::utils::getThreadID();
+
+    std::string message_id;
+    switch (getShowTimestampMode())
+    {
+        case 1: message_id = cv::format("%d@%0.3f", threadID, getTimestampNS() * 1e-9); break;
+        case 1+2: message_id = cv::format("%d@%llu", threadID, (long long unsigned int)getTimestampNS()); break;
+        default: message_id = cv::format("%d", threadID); break;
+    }
+
     std::ostringstream ss;
     switch (logLevel)
     {
-    case LOG_LEVEL_FATAL:   ss << "[FATAL:" << threadID << "] " << message << std::endl; break;
-    case LOG_LEVEL_ERROR:   ss << "[ERROR:" << threadID << "] " << message << std::endl; break;
-    case LOG_LEVEL_WARNING: ss << "[ WARN:" << threadID << "] " << message << std::endl; break;
-    case LOG_LEVEL_INFO:    ss << "[ INFO:" << threadID << "] " << message << std::endl; break;
-    case LOG_LEVEL_DEBUG:   ss << "[DEBUG:" << threadID << "] " << message << std::endl; break;
+    case LOG_LEVEL_FATAL:   ss << "[FATAL:" << message_id << "] " << message << std::endl; break;
+    case LOG_LEVEL_ERROR:   ss << "[ERROR:" << message_id << "] " << message << std::endl; break;
+    case LOG_LEVEL_WARNING: ss << "[ WARN:" << message_id << "] " << message << std::endl; break;
+    case LOG_LEVEL_INFO:    ss << "[ INFO:" << message_id << "] " << message << std::endl; break;
+    case LOG_LEVEL_DEBUG:   ss << "[DEBUG:" << message_id << "] " << message << std::endl; break;
     case LOG_LEVEL_VERBOSE: ss << message << std::endl; break;
     case LOG_LEVEL_SILENT: return;  // avoid compiler warning about incomplete switch
     case ENUM_LOG_LEVEL_FORCE_INT: return;  // avoid compiler warning about incomplete switch
diff --git a/modules/core/src/precomp.hpp b/modules/core/src/precomp.hpp
index 3057729928..d0582602b4 100644
--- a/modules/core/src/precomp.hpp
+++ b/modules/core/src/precomp.hpp
@@ -368,6 +368,10 @@ bool __termination;  // skip some cleanups, because process is terminating
 
 cv::Mutex& getInitializationMutex();
 
+/// @brief Returns timestamp in nanoseconds since program launch
+int64 getTimestampNS();
+
+
 #define CV_SINGLETON_LAZY_INIT_(TYPE, INITIALIZER, RET_VALUE) \
     static TYPE* const instance = INITIALIZER; \
     return RET_VALUE;
diff --git a/modules/core/src/system.cpp b/modules/core/src/system.cpp
index 3eb6f82e5f..3923f13b63 100644
--- a/modules/core/src/system.cpp
+++ b/modules/core/src/system.cpp
@@ -944,6 +944,51 @@ int64 getCPUTickCount(void)
 
 #endif
 
+
+namespace internal {
+
+class Timestamp
+{
+public:
+    const int64 zeroTickCount;
+    const double ns_in_ticks;
+
+    Timestamp()
+        : zeroTickCount(getTickCount())
+        , ns_in_ticks(1e9 / getTickFrequency())
+    {
+        // nothing
+    }
+
+    int64 getTimestamp()
+    {
+        int64 t = getTickCount();
+        return (int64)((t - zeroTickCount) * ns_in_ticks);
+    }
+
+    static Timestamp& getInstance()
+    {
+        static Timestamp g_timestamp;
+        return g_timestamp;
+    }
+};
+
+class InitTimestamp {
+public:
+    InitTimestamp() {
+        Timestamp::getInstance();
+    }
+};
+static InitTimestamp g_initialize_timestamp;  // force zero timestamp initialization
+
+}  // namespace
+
+int64 getTimestampNS()
+{
+    return internal::Timestamp::getInstance().getTimestamp();
+}
+
+
 const String& getBuildInformation()
 {
     static String build_info =
diff --git a/modules/core/src/trace.cpp b/modules/core/src/trace.cpp
index 1c12c2a2e5..1011db5e3b 100644
--- a/modules/core/src/trace.cpp
+++ b/modules/core/src/trace.cpp
@@ -63,15 +63,6 @@ namespace details {
 #pragma warning(disable:4065) // switch statement contains 'default' but no 'case' labels
 #endif
 
-static int64 g_zero_timestamp = 0;
-
-static int64 getTimestamp()
-{
-    int64 t = getTickCount();
-    static double tick_to_ns = 1e9 / getTickFrequency();
-    return (int64)((t - g_zero_timestamp) * tick_to_ns);
-}
-
 static bool getParameterTraceEnable()
 {
     static bool param_traceEnable = utils::getConfigurationParameterBool("OPENCV_TRACE", false);
@@ -485,7 +476,7 @@ Region::Region(const LocationStaticStorage& location) :
         }
     }
 
-    int64 beginTimestamp = getTimestamp();
+    int64 beginTimestamp = getTimestampNS();
 
     int currentDepth = ctx.getCurrentDepth() + 1;
     switch (location.flags & REGION_FLAG_IMPL_MASK)
@@ -635,7 +626,7 @@ void Region::destroy()
         }
     }
 
-    int64 endTimestamp = getTimestamp();
+    int64 endTimestamp = getTimestampNS();
     int64 duration = endTimestamp - ctx.stackTopBeginTimestamp();
 
     bool active = isActive();
@@ -844,7 +835,7 @@ static bool isInitialized = false;
 
 TraceManager::TraceManager()
 {
-    g_zero_timestamp = cv::getTickCount();
+    (void)cv::getTimestampNS();
 
     isInitialized = true;
     CV_LOG("TraceManager ctor: " << (void*)this);
@@ -990,7 +981,7 @@ void parallelForFinalize(const Region& rootRegion)
 {
     TraceManagerThreadLocal& ctx = getTraceManager().tls.getRef();
 
-    int64 endTimestamp = getTimestamp();
+    int64 endTimestamp = getTimestampNS();
     int64 duration = endTimestamp - ctx.stackTopBeginTimestamp();
     CV_LOG_PARALLEL(NULL, "parallel_for duration: " << duration << " " << &rootRegion);
 
diff --git a/modules/dnn/test/test_backends.cpp b/modules/dnn/test/test_backends.cpp
index e8c7e700f6..cefd22c824 100644
--- a/modules/dnn/test/test_backends.cpp
+++ b/modules/dnn/test/test_backends.cpp
@@ -217,8 +217,16 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe_Different_Width_Height)
 #if defined(INF_ENGINE_RELEASE)
     if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 || backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) &&
         target == DNN_TARGET_MYRIAD && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
-        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X);
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
 #endif
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE exception: Ngraph operation Transpose with name conv15_2_mbox_conf_perm has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
+
     Mat sample = imread(findDataFile("dnn/street.png"));
     Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 560), Scalar(127.5, 127.5, 127.5), false);
     float scoreDiff = 0.0, iouDiff = 0.0;
@@ -324,12 +332,14 @@ TEST_P(DNNTestNetwork, SSD_VGG16)
                  CV_TEST_TAG_DEBUG_VERYLONG);
     if (backend == DNN_BACKEND_HALIDE && target == DNN_TARGET_CPU)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_HALIDE);  // TODO HALIDE_CPU
+
     Mat sample = imread(findDataFile("dnn/street.png"));
     Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
+
     float scoreDiff = 0.0, iouDiff = 0.0;
     if (target == DNN_TARGET_OPENCL_FP16)
     {
-        scoreDiff = 0.0325;
+        scoreDiff = 0.04;
     }
     else if (target == DNN_TARGET_MYRIAD)
     {
diff --git a/modules/dnn/test/test_caffe_importer.cpp b/modules/dnn/test/test_caffe_importer.cpp
index 24e4fa21a2..e955443d59 100644
--- a/modules/dnn/test/test_caffe_importer.cpp
+++ b/modules/dnn/test/test_caffe_importer.cpp
@@ -517,10 +517,12 @@ TEST_P(Test_Caffe_nets, Colorization)
         l1 = 0.21;
         lInf = 4.5;
     }
+#if defined(INF_ENGINE_RELEASE)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
     {
-        l1 = 0.26; lInf = 6.5;
+        l1 = 0.3; lInf = 10;
     }
+#endif
 
     normAssert(out, ref, "", l1, lInf);
     expectNoFallbacksFromIE(net);
@@ -713,6 +715,13 @@ TEST_P(Test_Caffe_nets, FasterRCNN_zf)
 #endif
         CV_TEST_TAG_DEBUG_LONG
     );
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE exception: Ngraph operation Reshape with name rpn_cls_score_reshape has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
     if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
          backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_OPENCL_FP16)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
@@ -734,6 +743,11 @@ TEST_P(Test_Caffe_nets, RFCN)
         CV_TEST_TAG_LONG,
         CV_TEST_TAG_DEBUG_VERYLONG
     );
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // Exception: Function contains several inputs and outputs with one friendly name! (HETERO bug?)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
     if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
          backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_OPENCL_FP16)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
diff --git a/modules/dnn/test/test_common.hpp b/modules/dnn/test/test_common.hpp
index f20aa507c1..d7392c1bd9 100644
--- a/modules/dnn/test/test_common.hpp
+++ b/modules/dnn/test/test_common.hpp
@@ -172,16 +172,19 @@ public:
 
     static void checkBackend(int backend, int target, Mat* inp = 0, Mat* ref = 0)
     {
+        CV_UNUSED(backend); CV_UNUSED(target); CV_UNUSED(inp); CV_UNUSED(ref);
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021000000)
         if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 || backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
             && target == DNN_TARGET_MYRIAD)
         {
             if (inp && ref && inp->dims == 4 && ref->dims == 4 &&
                 inp->size[0] != 1 && inp->size[0] != ref->size[0])
             {
+                std::cout << "Inconsistent batch size of input and output blobs for Myriad plugin" << std::endl;
                 applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD);
-                throw SkipTestException("Inconsistent batch size of input and output blobs for Myriad plugin");
             }
         }
+#endif
     }
 
     void expectNoFallbacks(Net& net, bool raiseError = true)
diff --git a/modules/dnn/test/test_darknet_importer.cpp b/modules/dnn/test/test_darknet_importer.cpp
index d822efa889..21136416ce 100644
--- a/modules/dnn/test/test_darknet_importer.cpp
+++ b/modules/dnn/test/test_darknet_importer.cpp
@@ -245,13 +245,13 @@ public:
                 nms_boxes.push_back(box);
                 nms_confidences.push_back(conf);
                 nms_classIds.push_back(class_id);
-#if 0  // use to update test reference data
-                std::cout << b << ", " << class_id << ", " << conf << "f, "
-                          << box.x << "f, " << box.y << "f, "
-                          << box.x + box.width << "f, " << box.y + box.height << "f,"
-                          << std::endl;
-#endif
-
+                if (cvtest::debugLevel > 0)
+                {
+                    std::cout << b << ", " << class_id << ", " << conf << "f, "
+                              << box.x << "f, " << box.y << "f, "
+                              << box.x + box.width << "f, " << box.y + box.height << "f,"
+                              << std::endl;
+                }
             }
 
             if (cvIsNaN(iouDiff))
@@ -359,6 +359,13 @@ TEST_P(Test_Darknet_nets, YoloVoc)
         scoreDiff = 0.03;
         iouDiff = 0.018;
     }
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+    {
+        iouDiff = std::numeric_limits<double>::quiet_NaN();
+    }
+#endif
 
     std::string config_file = "yolo-voc.cfg";
     std::string weights_file = "yolo-voc.weights";
@@ -372,6 +379,12 @@ TEST_P(Test_Darknet_nets, YoloVoc)
     SCOPED_TRACE("batch size 2");
     testDarknetModel(config_file, weights_file, ref, scoreDiff, iouDiff, 0.24, nmsThreshold);
     }
+
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
 }
 
 TEST_P(Test_Darknet_nets, TinyYoloVoc)
@@ -615,6 +628,14 @@ TEST_P(Test_Darknet_nets, YOLOv4)
     std::string config_file = "yolov4.cfg";
     std::string weights_file = "yolov4.weights";
 
+
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy (batch 1)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+    {
+        iouDiff = std::numeric_limits<double>::quiet_NaN();
+    }
+#endif
 #if defined(INF_ENGINE_RELEASE)
     if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
          backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && target == DNN_TARGET_MYRIAD &&
@@ -633,6 +654,13 @@ TEST_P(Test_Darknet_nets, YOLOv4)
     {
         SCOPED_TRACE("batch size 2");
 
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy (batch 1)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+    {
+        iouDiff = 0.45f;
+    }
+#endif
 #if defined(INF_ENGINE_RELEASE)
         if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
         {
@@ -648,6 +676,12 @@ TEST_P(Test_Darknet_nets, YOLOv4)
 
         testDarknetModel(config_file, weights_file, ref, scoreDiff, iouDiff);
     }
+
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
 }
 
 TEST_P(Test_Darknet_nets, YOLOv4_tiny)
@@ -718,6 +752,13 @@ TEST_P(Test_Darknet_nets, YOLOv4x_mish)
 {
     applyTestTag(CV_TEST_TAG_LONG, (target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_1GB : CV_TEST_TAG_MEMORY_2GB));
 
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE exception: Ngraph operation Transpose with name permute_168 has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020040000)  // nGraph compilation failure
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
diff --git a/modules/dnn/test/test_halide_layers.cpp b/modules/dnn/test/test_halide_layers.cpp
index a6c1b7d0f8..0d0e313a06 100644
--- a/modules/dnn/test/test_halide_layers.cpp
+++ b/modules/dnn/test/test_halide_layers.cpp
@@ -39,12 +39,13 @@ static void test(Mat& input, Net& net, Backend backendId, Target targetId, bool
         l1 = default_l1;
     if (lInf == 0.0)
         lInf = default_lInf;
-#if 0
-    std::cout << "l1=" << l1 << "  lInf=" << lInf << std::endl;
-    std::cout << outputDefault.reshape(1, outputDefault.total()).t() << std::endl;
-    std::cout << outputHalide.reshape(1, outputDefault.total()).t() << std::endl;
-#endif
     normAssert(outputDefault, outputHalide, "", l1, lInf);
+    if (cvtest::debugLevel > 0 || testing::Test::HasFailure())
+    {
+        std::cout << "l1=" << l1 << "  lInf=" << lInf << std::endl;
+        std::cout << outputDefault.reshape(1, outputDefault.total()).t() << std::endl;
+        std::cout << outputHalide.reshape(1, outputDefault.total()).t() << std::endl;
+    }
 }
 
 static void test(LayerParams& params, Mat& input, Backend backendId, Target targetId, bool skipCheck = false, double l1 = 0.0, double lInf = 0.0)
@@ -802,6 +803,16 @@ TEST_P(Eltwise, Accuracy)
     Backend backendId = get<0>(get<4>(GetParam()));
     Target targetId = get<1>(get<4>(GetParam()));
 
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy
+    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && targetId == DNN_TARGET_OPENCL &&
+        inSize == Vec3i(1, 4, 5) && op == "sum" && numConv == 1 && !weighted)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+    if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && targetId == DNN_TARGET_OPENCL &&
+        inSize == Vec3i(2, 8, 6) && op == "sum" && numConv == 1 && !weighted)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
+
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LE(2018050000)
     if (backendId == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && targetId == DNN_TARGET_MYRIAD &&
         inSize == Vec3i(1, 4, 5))
diff --git a/modules/dnn/test/test_layers.cpp b/modules/dnn/test/test_layers.cpp
index ee0629b853..6b408f37c1 100644
--- a/modules/dnn/test/test_layers.cpp
+++ b/modules/dnn/test/test_layers.cpp
@@ -378,7 +378,7 @@ TEST_P(Test_Caffe_layers, layer_prelu_fc)
     // Reference output values are in range [-0.0001, 10.3906]
     double l1 = (target == DNN_TARGET_MYRIAD) ? 0.005 : 0.0;
     double lInf = (target == DNN_TARGET_MYRIAD) ? 0.021 : 0.0;
-#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020040000)
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020040000)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL)
     {
         l1 = 0.006f; lInf = 0.05f;
@@ -1475,6 +1475,14 @@ TEST_P(Test_DLDT_two_inputs, as_backend)
         lInf = 0.3;
     }
     normAssert(out, ref, "", l1, lInf);
+    if (cvtest::debugLevel > 0 || HasFailure())
+    {
+        std::cout << "input1 scale=" << kScale << " input2 scale=" << kScaleInv << std::endl;
+        std::cout << "input1: " << firstInp.size << " " << firstInp.reshape(1, 1) << std::endl;
+        std::cout << "input2: " << secondInp.size << " " << secondInp.reshape(1, 1) << std::endl;
+        std::cout << "ref: " << ref.reshape(1, 1) << std::endl;
+        std::cout << "out: " << out.reshape(1, 1) << std::endl;
+    }
 }
 
 INSTANTIATE_TEST_CASE_P(/*nothing*/, Test_DLDT_two_inputs, Combine(
diff --git a/modules/dnn/test/test_onnx_importer.cpp b/modules/dnn/test/test_onnx_importer.cpp
index 56a6ab6ea1..f5696e54dd 100644
--- a/modules/dnn/test/test_onnx_importer.cpp
+++ b/modules/dnn/test/test_onnx_importer.cpp
@@ -802,6 +802,14 @@ TEST_P(Test_ONNX_layers, Split_EltwiseMax)
 
 TEST_P(Test_ONNX_layers, LSTM_Activations)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE Exception: Ngraph operation Reshape with name Block1237_Output_0_before_reshape has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
+
     testONNXModels("lstm_cntk_tanh", pb, 0, 0, false, false);
 }
 
@@ -946,6 +954,13 @@ TEST_P(Test_ONNX_layers, Conv1d_variable_weight_bias)
 
 TEST_P(Test_ONNX_layers, GatherMultiOutput)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE Exception: Ngraph operation Reshape with name 6 has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021030000)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // exception
@@ -953,7 +968,7 @@ TEST_P(Test_ONNX_layers, GatherMultiOutput)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // exception
 #endif
 
-#if defined(INF_ENGINE_RELEASE)
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LE(2021030000)
     if (target == DNN_TARGET_MYRIAD)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE);
 #endif
@@ -963,14 +978,25 @@ TEST_P(Test_ONNX_layers, GatherMultiOutput)
 
 TEST_P(Test_ONNX_layers, DynamicAxes)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // accuracy
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
+#if defined(INF_ENGINE_RELEASE)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
     {
         if (target == DNN_TARGET_MYRIAD) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
     }
+#if INF_ENGINE_VER_MAJOR_LT(2021000000)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
     {
         if (target == DNN_TARGET_MYRIAD) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
     }
+#endif
+#endif
     testONNXModels("squeeze_and_conv_dynamic_axes");
     testONNXModels("unsqueeze_and_conv_dynamic_axes");
     testONNXModels("gather_dynamic_axes");
@@ -1050,6 +1076,13 @@ TEST_P(Test_ONNX_layers, PoolConv1d)
 
 TEST_P(Test_ONNX_layers, ConvResizePool1d)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE Exception: Ngraph operation Reshape with name 15 has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
 #if defined(INF_ENGINE_RELEASE)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
     {
@@ -1367,8 +1400,8 @@ TEST_P(Test_ONNX_nets, TinyYolov2)
     double l1 =  default_l1, lInf = default_lInf;
     if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
     {
-        l1 = 0.017;
-        lInf = 0.14;
+        l1 = 0.02;
+        lInf = 0.2;
     }
     else if (target == DNN_TARGET_CUDA_FP16)
     {
@@ -1465,10 +1498,10 @@ TEST_P(Test_ONNX_nets, Emotion_ferplus)
         l1 = 2.4e-4;
         lInf = 6e-4;
     }
-#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020040000)
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020040000)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
     {
-        l1 = 0.012f; lInf = 0.035f;
+        l1 = 0.013f; lInf = 0.035f;
     }
 #endif
 
diff --git a/modules/dnn/test/test_tf_importer.cpp b/modules/dnn/test/test_tf_importer.cpp
index 812523311b..ed8e8648e0 100644
--- a/modules/dnn/test/test_tf_importer.cpp
+++ b/modules/dnn/test/test_tf_importer.cpp
@@ -84,6 +84,10 @@ public:
     void runTensorFlowNet(const std::string& prefix, bool hasText = false,
                           double l1 = 0.0, double lInf = 0.0, bool memoryLoad = false, const std::string& groupPrefix = "")
     {
+        if (cvtest::debugLevel > 0)
+        {
+            std::cout << prefix << groupPrefix << std::endl;
+        }
         std::string netPath = path(prefix + groupPrefix + "_net.pb");
         std::string netConfig = (hasText ? path(prefix + groupPrefix + "_net.pbtxt") : "");
         std::string inpPath = path(prefix + "_in.npy");
@@ -119,6 +123,16 @@ public:
         net.setInput(input);
         cv::Mat output = net.forward();
         normAssert(ref, output, "", l1 ? l1 : default_l1, lInf ? lInf : default_lInf);
+
+        if (cvtest::debugLevel > 0 || HasFailure())
+        {
+            std::cout << "input: " << input.size << std::endl;
+            std::cout << input.reshape(1, 1) << std::endl;
+            std::cout << "ref " << ref.size << std::endl;
+            std::cout << ref.reshape(1, 1) << std::endl;
+            std::cout << "output: " << output.size << std::endl;
+            std::cout << output.reshape(1, 1) << std::endl;
+        }
     }
 };
 
@@ -133,7 +147,7 @@ TEST_P(Test_TensorFlow_layers, reduce_max)
 {
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
-    runTensorFlowNet("max_pool_by_axis");
+    runTensorFlowNet("max_pool_by_axis", false, 0.0f, 0.0f);
 }
 
 TEST_P(Test_TensorFlow_layers, reduce_sum)
@@ -145,7 +159,11 @@ TEST_P(Test_TensorFlow_layers, reduce_sum)
 
 TEST_P(Test_TensorFlow_layers, reduce_max_channel)
 {
-    runTensorFlowNet("reduce_max_channel");
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020040000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)  // incorrect result
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
+    runTensorFlowNet("reduce_max_channel", false, 0.0f, 0.0f);
 }
 
 TEST_P(Test_TensorFlow_layers, reduce_sum_channel)
@@ -155,6 +173,10 @@ TEST_P(Test_TensorFlow_layers, reduce_sum_channel)
 
 TEST_P(Test_TensorFlow_layers, reduce_max_channel_keep_dims)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020040000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)  // incorrect result
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
     runTensorFlowNet("reduce_max_channel", false, 0.0, 0.0, false, "_keep_dims");
 }
 
@@ -221,13 +243,49 @@ TEST_P(Test_TensorFlow_layers, padding)
     runTensorFlowNet("keras_pad_concat");
 }
 
-TEST_P(Test_TensorFlow_layers, padding_asymmetric)
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_1)
 {
     runTensorFlowNet("conv2d_asymmetric_pads_nchw");
+}
+
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_2)
+{
     runTensorFlowNet("conv2d_asymmetric_pads_nhwc");
+}
+
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_3)
+{
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021040000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_CPU)  // Exception: Unsupported pad value
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_CPU, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020020000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)  // Exception: Unsupported pad value
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
     runTensorFlowNet("max_pool2d_asymmetric_pads_nchw");
+}
+
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_4)
+{
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021040000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_CPU)  // Exception: Unsupported pad value
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_CPU, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2020020000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)  // Exception: Unsupported pad value
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
     runTensorFlowNet("max_pool2d_asymmetric_pads_nhwc");
+}
+
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_5)
+{
     runTensorFlowNet("conv2d_backprop_input_asymmetric_pads_nchw");
+}
+
+TEST_P(Test_TensorFlow_layers, padding_asymmetric_6)
+{
     runTensorFlowNet("conv2d_backprop_input_asymmetric_pads_nhwc");
 }
 
@@ -268,6 +326,13 @@ TEST_P(Test_TensorFlow_layers, pad_and_concat)
 
 TEST_P(Test_TensorFlow_layers, concat_axis_1)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE Exception: Ngraph operation Transpose with name Flatten_1/flatten/Reshape/nhwc has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021030000)
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);  // exception
@@ -423,19 +488,77 @@ TEST_P(Test_TensorFlow_layers, pooling_reduce_sum)
     runTensorFlowNet("reduce_sum");  // a SUM pooling over all spatial dimensions.
 }
 
-TEST_P(Test_TensorFlow_layers, pooling_reduce_sum2)
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_0_false)
 {
-    int axises[] = {0, 1, 2, 3};
-    for (int keepdims = 0; keepdims <= 1; ++keepdims)
-    {
-        for (int i = 0; i < sizeof(axises)/sizeof(axises[0]); ++i)
-        {
-            runTensorFlowNet(cv::format("reduce_sum_%d_%s", axises[i], (keepdims ? "True" : "False")));
-        }
-        runTensorFlowNet(cv::format("reduce_sum_1_2_%s", keepdims ? "True" : "False"));
-    }
+    runTensorFlowNet("reduce_sum_0_False");
 }
 
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_1_false)
+{
+    runTensorFlowNet("reduce_sum_1_False");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_2_false)
+{
+    runTensorFlowNet("reduce_sum_2_False");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_3_false)
+{
+    runTensorFlowNet("reduce_sum_3_False");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_1_2_false)
+{
+#if defined(INF_ENGINE_RELEASE)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
+    {
+        default_l1 = 0.01f;
+    }
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+    {
+        default_l1 = 0.01f;
+    }
+#endif
+    runTensorFlowNet("reduce_sum_1_2_False");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_0_true)
+{
+    runTensorFlowNet("reduce_sum_0_True");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_1_true)
+{
+    runTensorFlowNet("reduce_sum_1_True");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_2_true)
+{
+    runTensorFlowNet("reduce_sum_2_True");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_3_true)
+{
+    runTensorFlowNet("reduce_sum_3_True");
+}
+
+TEST_P(Test_TensorFlow_layers, pooling_reduce_sum_1_2_true)
+{
+#if defined(INF_ENGINE_RELEASE)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
+    {
+        default_l1 = 0.01f;
+    }
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+    {
+        default_l1 = 0.01f;
+    }
+#endif
+    runTensorFlowNet("reduce_sum_1_2_True");
+}
+
+
 TEST_P(Test_TensorFlow_layers, max_pool_grad)
 {
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
@@ -715,13 +838,14 @@ TEST_P(Test_TensorFlow_nets, MobileNet_SSD)
     double scoreDiff = default_l1, iouDiff = default_lInf;
     if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
     {
-        scoreDiff = 0.0043;
-        iouDiff = 0.037;
+        scoreDiff = 0.01;
+        iouDiff = 0.1;
     }
     else if (target == DNN_TARGET_CUDA_FP16)
     {
         iouDiff = 0.04;
     }
+
     normAssertDetections(ref, out, "", 0.2, scoreDiff, iouDiff);
 #if defined(INF_ENGINE_RELEASE) && INF_ENGINE_RELEASE >= 2019010000
     expectNoFallbacksFromIE(net);
@@ -815,16 +939,13 @@ TEST_P(Test_TensorFlow_nets, MobileNet_v1_SSD)
     expectNoFallbacksFromIE(net);
 }
 
-TEST_P(Test_TensorFlow_nets, Faster_RCNN)
+TEST_P(Test_TensorFlow_nets, Faster_RCNN_inception_v2_coco_2018_01_28)
 {
-    // FIXIT split test
     applyTestTag(
         (target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_1GB : CV_TEST_TAG_MEMORY_2GB),
         CV_TEST_TAG_LONG,
         CV_TEST_TAG_DEBUG_VERYLONG
     );
-    static std::string names[] = {"faster_rcnn_inception_v2_coco_2018_01_28",
-                                  "faster_rcnn_resnet50_coco_2018_01_28"};
 
 #ifdef INF_ENGINE_RELEASE
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 &&
@@ -835,13 +956,82 @@ TEST_P(Test_TensorFlow_nets, Faster_RCNN)
         backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
 #endif
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021040000)
     // segfault: inference-engine/thirdparty/clDNN/src/gpu/detection_output_cpu.cpp:111:
     // Assertion `prior_height > 0' failed.
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
+#endif
 
-    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
-        applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
+    if (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
+
+    checkBackend();
+
+    double scoresDiff = 1e-5;
+    double iouDiff = 1e-4;
+
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 || backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
+    {
+        scoresDiff = 0.02;
+        iouDiff = 0.1;
+    }
+
+    std::string name = "faster_rcnn_inception_v2_coco_2018_01_28";
+    {
+        std::string proto = findDataFile("dnn/" + name + ".pbtxt");
+        std::string model = findDataFile("dnn/" + name + ".pb", false);
+
+        Net net = readNetFromTensorflow(model, proto);
+        net.setPreferableBackend(backend);
+        net.setPreferableTarget(target);
+        Mat img = imread(findDataFile("dnn/dog416.png"));
+        Mat blob = blobFromImage(img, 1.0f, Size(800, 600), Scalar(), true, false);
+
+        net.setInput(blob);
+        Mat out = net.forward();
+
+        Mat ref = blobFromNPY(findDataFile("dnn/tensorflow/" + name + ".detection_out.npy"));
+
+        // accuracy (both OpenCV & IE)
+        if (target == DNN_TARGET_OPENCL_FP16)
+            applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
+
+        normAssertDetections(ref, out, name.c_str(), 0.3, scoresDiff, iouDiff);
+    }
+}
+
+TEST_P(Test_TensorFlow_nets, Faster_RCNN_resnet50_coco_2018_01_28)
+{
+    applyTestTag(
+        (target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_1GB : CV_TEST_TAG_MEMORY_2GB),
+        CV_TEST_TAG_LONG,
+        CV_TEST_TAG_DEBUG_VERYLONG
+    );
+
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    // IE exception: Ngraph operation Transpose with name FirstStageBoxPredictor/ClassPredictor/reshape_1/nhwc has dynamic output shape on 0 port, but CPU plug-in supports only static shape
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && (target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
+        applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
+            CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION
+        );
+#endif
+
+#ifdef INF_ENGINE_RELEASE
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 &&
+        (INF_ENGINE_VER_MAJOR_LT(2019020000) || target != DNN_TARGET_CPU))
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+
+    if (INF_ENGINE_VER_MAJOR_GT(2019030000) &&
+        backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
+#endif
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_LT(2021040000)
+    // segfault: inference-engine/thirdparty/clDNN/src/gpu/detection_output_cpu.cpp:111:
+    // Assertion `prior_height > 0' failed.
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
+#endif
 
     if (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
@@ -856,10 +1046,11 @@ TEST_P(Test_TensorFlow_nets, Faster_RCNN)
         scoresDiff = 0.06;
         iouDiff = 0.08;
     }
-    for (int i = 0; i < 2; ++i)
+
+    std::string name = "faster_rcnn_resnet50_coco_2018_01_28";
     {
-        std::string proto = findDataFile("dnn/" + names[i] + ".pbtxt");
-        std::string model = findDataFile("dnn/" + names[i] + ".pb", false);
+        std::string proto = findDataFile("dnn/" + name + ".pbtxt");
+        std::string model = findDataFile("dnn/" + name + ".pb", false);
 
         Net net = readNetFromTensorflow(model, proto);
         net.setPreferableBackend(backend);
@@ -870,8 +1061,13 @@ TEST_P(Test_TensorFlow_nets, Faster_RCNN)
         net.setInput(blob);
         Mat out = net.forward();
 
-        Mat ref = blobFromNPY(findDataFile("dnn/tensorflow/" + names[i] + ".detection_out.npy"));
-        normAssertDetections(ref, out, names[i].c_str(), 0.3, scoresDiff, iouDiff);
+        Mat ref = blobFromNPY(findDataFile("dnn/tensorflow/" + name + ".detection_out.npy"));
+
+        // accuracy
+        if (target == DNN_TARGET_OPENCL_FP16)
+            applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
+
+        normAssertDetections(ref, out, name.c_str(), 0.3, scoresDiff, iouDiff);
     }
 }
 
@@ -1282,6 +1478,10 @@ TEST_P(Test_TensorFlow_layers, resize_bilinear_down)
 
 TEST_P(Test_TensorFlow_layers, resize_concat_optimization)
 {
+#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021040000)
+    if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)  // Exception: Function contains several inputs and outputs with one friendly name! (HETERO bug?)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
+#endif
     runTensorFlowNet("resize_concat_optimization");
 }
 
@@ -1406,7 +1606,7 @@ TEST_P(Test_TensorFlow_nets, Mask_RCNN)
     Mat outDetections = outs[0];
     Mat outMasks = outs[1];
 
-    double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.019 : 2e-5;
+    double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.2 : 2e-5;
     double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.018 : default_lInf;
     normAssertDetections(refDetections, outDetections, "", /*threshold for zero confidence*/1e-5, scoreDiff, iouDiff);
 
@@ -1440,7 +1640,7 @@ TEST_P(Test_TensorFlow_nets, Mask_RCNN)
 
     double inter = cv::countNonZero(masks & refMasks);
     double area = cv::countNonZero(masks | refMasks);
-    EXPECT_GE(inter / area, 0.99);
+    EXPECT_GE(inter / area, (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.98 : 0.99);
 
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
         expectNoFallbacks(net);
diff --git a/modules/dnn/test/test_torch_importer.cpp b/modules/dnn/test/test_torch_importer.cpp
index 48725f244e..49b593caa6 100644
--- a/modules/dnn/test/test_torch_importer.cpp
+++ b/modules/dnn/test/test_torch_importer.cpp
@@ -186,7 +186,7 @@ TEST_P(Test_Torch_layers, run_concat)
 TEST_P(Test_Torch_layers, run_depth_concat)
 {
     double lInf = 0.0;
-    if (target == DNN_TARGET_OPENCL_FP16)
+    if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
     {
         lInf = 0.032;
     }
diff --git a/modules/highgui/CMakeLists.txt b/modules/highgui/CMakeLists.txt
index 5c0f1e3278..9177c1ba46 100644
--- a/modules/highgui/CMakeLists.txt
+++ b/modules/highgui/CMakeLists.txt
@@ -66,7 +66,7 @@ if(HAVE_QT)
       QT5_ADD_RESOURCES(_RCC_OUTFILES ${CMAKE_CURRENT_LIST_DIR}/src/window_QT.qrc)
       QT5_WRAP_CPP(_MOC_OUTFILES ${CMAKE_CURRENT_LIST_DIR}/src/window_QT.h)
     else()
-      message(FATAL_ERROR "Unsuported QT version: ${QT_VERSION_MAJOR}")
+      message(FATAL_ERROR "Unsupported QT version: ${QT_VERSION_MAJOR}")
     endif()
 
     list(APPEND highgui_srcs
diff --git a/modules/js/generator/embindgen.py b/modules/js/generator/embindgen.py
index 2ad6479ba5..f9b8431402 100644
--- a/modules/js/generator/embindgen.py
+++ b/modules/js/generator/embindgen.py
@@ -485,7 +485,7 @@ class JSWrapperGenerator(object):
                 arg_types.append(arg_type)
                 unwrapped_arg_types.append(arg_type)
 
-            # Function attribure
+            # Function attribute
             func_attribs = ''
             if '*' in ''.join(arg_types):
                 func_attribs += ', allow_raw_pointers()'
@@ -680,7 +680,7 @@ class JSWrapperGenerator(object):
                     def_args.append(arg.defval)
                 arg_types.append(orig_arg_types[-1])
 
-            # Function attribure
+            # Function attribute
             func_attribs = ''
             if '*' in ''.join(orig_arg_types):
                 func_attribs += ', allow_raw_pointers()'
diff --git a/modules/ts/misc/chart.py b/modules/ts/misc/chart.py
index 0547eefdfc..6ae726abeb 100755
--- a/modules/ts/misc/chart.py
+++ b/modules/ts/misc/chart.py
@@ -1,5 +1,6 @@
 #!/usr/bin/env python
 
+from __future__ import print_function
 import testlog_parser, sys, os, xml, re
 from table_formatter import *
 from optparse import OptionParser
@@ -116,7 +117,7 @@ if __name__ == "__main__":
     (options, args) = parser.parse_args()
 
     if len(args) != 1:
-        print >> sys.stderr, "Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml"
+        print("Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml", file=sys.stderr)
         exit(1)
 
     options.generateHtml = detectHtmlOutputType(options.format)
@@ -136,7 +137,7 @@ if __name__ == "__main__":
     args[0] = os.path.basename(args[0])
 
     if not tests:
-        print >> sys.stderr, "Error - no tests matched"
+        print("Error - no tests matched", file=sys.stderr)
         exit(1)
 
     argsnum = len(tests[0][1])
@@ -156,26 +157,26 @@ if __name__ == "__main__":
             names1.add(sn)
         if sn == sname:
             if len(pair[1]) != argsnum:
-                print >> sys.stderr, "Error - unable to create chart tables for functions having different argument numbers"
+                print("Error - unable to create chart tables for functions having different argument numbers", file=sys.stderr)
                 sys.exit(1)
             for i in range(argsnum):
                 arglists[i][pair[1][i]] = 1
 
     if names1 or len(names) != 1:
-        print >> sys.stderr, "Error - unable to create tables for functions from different test suits:"
+        print("Error - unable to create tables for functions from different test suits:", file=sys.stderr)
         i = 1
         for name in sorted(names):
-            print >> sys.stderr, "%4s:   %s" % (i, name)
+            print("%4s:   %s" % (i, name), file=sys.stderr)
             i += 1
         if names1:
-            print >> sys.stderr, "Other suits in this log (can not be chosen):"
+            print("Other suits in this log (can not be chosen):", file=sys.stderr)
             for name in sorted(names1):
-                print >> sys.stderr, "%4s:   %s" % (i, name)
+                print("%4s:   %s" % (i, name), file=sys.stderr)
                 i += 1
         sys.exit(1)
 
     if argsnum < 2:
-        print >> sys.stderr, "Error - tests from %s have less than 2 parameters" % sname
+        print("Error - tests from %s have less than 2 parameters" % sname, file=sys.stderr)
         exit(1)
 
     for i in range(argsnum):
diff --git a/modules/ts/misc/report.py b/modules/ts/misc/report.py
index f0a87f2a12..ec60e16a13 100755
--- a/modules/ts/misc/report.py
+++ b/modules/ts/misc/report.py
@@ -1,5 +1,6 @@
 #!/usr/bin/env python
 
+from __future__ import print_function
 import testlog_parser, sys, os, xml, re, glob
 from table_formatter import *
 from optparse import OptionParser
@@ -14,7 +15,7 @@ if __name__ == "__main__":
     (options, args) = parser.parse_args()
 
     if len(args) < 1:
-        print >> sys.stderr, "Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml"
+        print("Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml", file=sys.stderr)
         exit(0)
 
     options.generateHtml = detectHtmlOutputType(options.format)
diff --git a/modules/ts/misc/run.py b/modules/ts/misc/run.py
index c2e4d6532b..bb17aa884f 100755
--- a/modules/ts/misc/run.py
+++ b/modules/ts/misc/run.py
@@ -105,7 +105,7 @@ if __name__ == "__main__":
     path = args.build_path
     try:
         if not os.path.isdir(path):
-            raise Err("Not a directory (should contain CMakeCache.txt ot test executables)")
+            raise Err("Not a directory (should contain CMakeCache.txt to test executables)")
         cache = CMakeCache(args.configuration)
         fname = os.path.join(path, "CMakeCache.txt")
 
diff --git a/modules/ts/misc/summary.py b/modules/ts/misc/summary.py
index 9da1fb60c6..bfb98358d2 100755
--- a/modules/ts/misc/summary.py
+++ b/modules/ts/misc/summary.py
@@ -1,5 +1,6 @@
 #!/usr/bin/env python
 
+from __future__ import print_function
 import testlog_parser, sys, os, xml, glob, re
 from table_formatter import *
 from optparse import OptionParser
@@ -26,7 +27,7 @@ def getSetName(tset, idx, columns, short = True):
 
 if __name__ == "__main__":
     if len(sys.argv) < 2:
-        print >> sys.stderr, "Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml [<log_name2>.xml ...]"
+        print("Usage:\n", os.path.basename(sys.argv[0]), "<log_name1>.xml [<log_name2>.xml ...]", file=sys.stderr)
         exit(0)
 
     parser = OptionParser()
diff --git a/platforms/winpack_dldt/build_package.py b/platforms/winpack_dldt/build_package.py
index db5d75c17d..36e86792cf 100644
--- a/platforms/winpack_dldt/build_package.py
+++ b/platforms/winpack_dldt/build_package.py
@@ -163,7 +163,7 @@ class BuilderDLDT:
         self.config = config
 
         cpath = self.config.dldt_config
-        log.info('DLDT build configration: %s', cpath)
+        log.info('DLDT build configuration: %s', cpath)
         if not os.path.exists(cpath):
             cpath = os.path.join(SCRIPT_DIR, cpath)
             if not os.path.exists(cpath):
@@ -575,5 +575,5 @@ if __name__ == "__main__":
     try:
         main()
     except:
-        log.info('FATAL: Error occured. To investigate problem try to change logging level using LOGLEVEL=DEBUG environment variable.')
+        log.info('FATAL: Error occurred. To investigate problem try to change logging level using LOGLEVEL=DEBUG environment variable.')
         raise
diff --git a/samples/dnn/siamrpnpp.py b/samples/dnn/siamrpnpp.py
index 2e15ec6708..0247242c62 100644
--- a/samples/dnn/siamrpnpp.py
+++ b/samples/dnn/siamrpnpp.py
@@ -300,7 +300,7 @@ class SiamRPNTracker:
         # clip boundary
         cx, cy, width, height = self._bbox_clip(cx, cy, width, height, img.shape[:2])
 
-        # udpate state
+        # update state
         self.center_pos = np.array([cx, cy])
         self.w = width
         self.h = height
diff --git a/samples/dnn/tf_text_graph_ssd.py b/samples/dnn/tf_text_graph_ssd.py
index dbdee6ea9f..d27fd0d384 100644
--- a/samples/dnn/tf_text_graph_ssd.py
+++ b/samples/dnn/tf_text_graph_ssd.py
@@ -270,12 +270,12 @@ def createSSDGraph(modelPath, configPath, outputPath):
     addConstNode('concat/axis_flatten', [-1], graph_def)
     addConstNode('PriorBox/concat/axis', [-2], graph_def)
 
-    for label in ['ClassPredictor', 'BoxEncodingPredictor' if box_predictor is 'convolutional' else 'BoxPredictor']:
+    for label in ['ClassPredictor', 'BoxEncodingPredictor' if box_predictor == 'convolutional' else 'BoxPredictor']:
         concatInputs = []
         for i in range(num_layers):
             # Flatten predictions
             flatten = NodeDef()
-            if box_predictor is 'convolutional':
+            if box_predictor == 'convolutional':
                 inpName = 'BoxPredictor_%d/%s/BiasAdd' % (i, label)
             else:
                 if i == 0:
@@ -308,7 +308,7 @@ def createSSDGraph(modelPath, configPath, outputPath):
         priorBox = NodeDef()
         priorBox.name = 'PriorBox_%d' % i
         priorBox.op = 'PriorBox'
-        if box_predictor is 'convolutional':
+        if box_predictor == 'convolutional':
             priorBox.input.append('BoxPredictor_%d/BoxEncodingPredictor/BiasAdd' % i)
         else:
             if i == 0:
diff --git a/samples/python/tutorial_code/video/meanshift/camshift.py b/samples/python/tutorial_code/video/meanshift/camshift.py
index d115bdb4d3..774e8b6c0e 100644
--- a/samples/python/tutorial_code/video/meanshift/camshift.py
+++ b/samples/python/tutorial_code/video/meanshift/camshift.py
@@ -24,7 +24,7 @@ mask = cv.inRange(hsv_roi, np.array((0., 60.,32.)), np.array((180.,255.,255.)))
 roi_hist = cv.calcHist([hsv_roi],[0],mask,[180],[0,180])
 cv.normalize(roi_hist,roi_hist,0,255,cv.NORM_MINMAX)
 
-# Setup the termination criteria, either 10 iteration or move by atleast 1 pt
+# Setup the termination criteria, either 10 iteration or move by at least 1 pt
 term_crit = ( cv.TERM_CRITERIA_EPS | cv.TERM_CRITERIA_COUNT, 10, 1 )
 
 while(1):
diff --git a/samples/python/tutorial_code/video/meanshift/meanshift.py b/samples/python/tutorial_code/video/meanshift/meanshift.py
index f765b023e9..ecdd573cb1 100644
--- a/samples/python/tutorial_code/video/meanshift/meanshift.py
+++ b/samples/python/tutorial_code/video/meanshift/meanshift.py
@@ -24,7 +24,7 @@ mask = cv.inRange(hsv_roi, np.array((0., 60.,32.)), np.array((180.,255.,255.)))
 roi_hist = cv.calcHist([hsv_roi],[0],mask,[180],[0,180])
 cv.normalize(roi_hist,roi_hist,0,255,cv.NORM_MINMAX)
 
-# Setup the termination criteria, either 10 iteration or move by atleast 1 pt
+# Setup the termination criteria, either 10 iteration or move by at least 1 pt
 term_crit = ( cv.TERM_CRITERIA_EPS | cv.TERM_CRITERIA_COUNT, 10, 1 )
 
 while(1):