OpenCV can change its behavior depending on the runtime environment:
- enable extra debugging output or performance tracing
- modify default locations and search paths
- tune some algorithms or general behavior
- enable or disable workarounds, safety features and optimizations
**Notes:**
- ⭐ marks most popular variables
- variables with names like this `VAR_${NAME}` describes family of variables, where `${NAME}` should be changed to one of predefined values, e.g. `VAR_TBB`, `VAR_OPENMP`, ...
@note This method may not work on all operating systems and/or Python distributions. For example, it works on Ubuntu Linux with system Python interpreter, but doesn't work on Windows 10 with the official Python package. It depends on the ability of a process to change its own environment (OpenCV uses `getenv` from C++ runtime to read variables).
Some modules have multiple available backends, following variables allow choosing specific backend or changing default priorities in which backends will be probed (e.g. when opening a video file).
| name | type | default | description |
|------|------|---------|-------------|
| OPENCV_PARALLEL_BACKEND | string | | choose specific paralel_for backend (one of `TBB`, `ONETBB`, `OPENMP`) |
| OPENCV_PARALLEL_PRIORITY_${NAME} | num | | set backend priority, default is 1000 |
| OPENCV_PARALLEL_PRIORITY_LIST | string, `,`-separated | | list of backends in priority order |
| OPENCV_UI_BACKEND | string | | choose highgui backend for window rendering (one of `GTK`, `GTK3`, `GTK2`, `QT`, `WIN32`) |
| OPENCV_UI_PRIORITY_${NAME} | num | | set highgui backend priority, default is 1000 |
Some external dependencies can be detached into a dynamic library, which will be loaded at runtime (plugin). Following variables allow changing default search locations and naming pattern for these plugins.
| name | type | default | description |
|------|------|---------|-------------|
| OPENCV_CORE_PLUGIN_PATH | paths | | directories to search for _core_ plugins |
| OPENCV_CORE_PARALLEL_PLUGIN_${NAME} | string, glob | | parallel_for plugin library name (glob), e.g. default for TBB is "opencv_core_parallel_tbb*.so" |
| OPENCV_DNN_PLUGIN_PATH | paths | | directories to search for _dnn_ plugins |
| OPENCV_DNN_PLUGIN_${NAME} | string, glob | | parallel_for plugin library name (glob), e.g. default for TBB is "opencv_core_parallel_tbb*.so" |
| OPENCV_CORE_PLUGIN_PATH | paths | | directories to search for _highgui_ plugins (YES it is CORE) |
**Note:** OpenCL device specification format is `<Platform>:<CPU|GPU|ACCELERATOR|nothing=GPU/CPU>:<deviceName>`, e.g. `AMD:GPU:`
| name | type | default | description |
|------|------|---------|-------------|
| OPENCV_OPENCL_RUNTIME | filepath or `disabled` | | path to OpenCL runtime library (e.g. `OpenCL.dll`, `libOpenCL.so`) |
| ⭐ OPENCV_OPENCL_DEVICE | string or `disabled` | | choose specific OpenCL device. See specification format in the note above. See more details in the Links section. |
| OPENCV_OPENCL_RAISE_ERROR | bool | false | raise exception if something fails during OpenCL kernel preparation and execution (Release builds only) |
| OPENCV_OPENCL_CACHE_WRITE | bool | true | allow writing to the cache, otherwise cache will be read-only |
| OPENCV_OPENCL_CACHE_LOCK_ENABLE | bool | true | use .lock files to synchronize between multiple applications using the same OpenCL cache (may not work on network drives) |
| OPENCV_OPENCL_CACHE_CLEANUP | bool | true | automatically remove old entries from cache (leftovers from older OpenCL runtimes) |
| OPENCV_OPENCL_BUFFERPOOL_LIMIT | num | 1 <<27(Inteldevice),0(others)|limitmemoryusedbybufferbool|
| OPENCV_OPENCL_HOST_PTR_BUFFERPOOL_LIMIT | num | | same as OPENCV_OPENCL_BUFFERPOOL_LIMIT, but for HOST_PTR buffers |
| OPENCV_OPENCL_BUFFER_FORCE_MAPPING | bool | false | force clEnqueueMapBuffer |
| OPENCV_OPENCL_BUFFER_FORCE_COPYING | bool | false | force clEnqueueReadBuffer/clEnqueueWriteBuffer |
| OPENCV_OPENCL_FORCE | bool | false | force running OpenCL kernel even if usual conditions are not met (e.g. dst.isUMat) |
| OPENCV_OPENCL_PERF_CHECK_BYPASS | bool | false | force running OpenCL kernel even if usual performance-related conditions are not met (e.g. image is very small) |
**Note:** In the table below `dump_base_name` equals to `ocv_dnn_net_%05d_%02d` where first argument is internal network ID and the second - dump level.
| name | type | default | description |
|------|------|---------|-------------|
| OPENCV_DNN_BACKEND_DEFAULT | num | 3 (OpenCV) | set default DNN backend, see dnn.hpp for backends enumeration |
| OPENCV_DNN_NETWORK_DUMP | num | 0 | level of information dumps, 0 - no dumps (default file name `${dump_base_name}.dot`) |
| OPENCV_DNN_OPENCL_ALLOW_ALL_DEVICES | bool | false | allow running on CPU devices, allow FP16 on non-Intel device |
| OPENCV_OCL4DNN_CONVOLUTION_IGNORE_INPUT_DIMS_4_CHECK | bool | false | workaround for OpenCL backend, see https://github.com/opencv/opencv/issues/20833 |
| OPENCV_OCL4DNN_WORKAROUND_IDLF | bool | true | another workaround for OpenCL backend |
| OPENCV_OCL4DNN_CONFIG_PATH | path | | path to kernel configuration cache for auto-tuning (must be existing directory), set this variable to enable auto-tuning |
| ⭐ OPENCV_TEST_DATA_PATH | dir path | | set test data search location (e.g. `/home/user/opencv_extra/testdata`) |
| ⭐ OPENCV_DNN_TEST_DATA_PATH | dir path | `$OPENCV_TEST_DATA_PATH/dnn` | set DNN model search location for tests (used by _dnn_, _gapi_, _objdetect_, _video_ modules) |
| OPENCV_OPEN_MODEL_ZOO_DATA_PATH | dir path | `$OPENCV_DNN_TEST_DATA_PATH/omz_intel_models` | set OpenVINO models search location for tests (used by _dnn_, _gapi_ modules) |
| INTEL_CVSDK_DIR | | | some _dnn_ tests can search OpenVINO models here too |
| OPENCV_TEST_DEBUG | num | 0 | debug level for tests, same as `--test_debug` (0 - no debug (default), 1 - basic test debug information, >1 - extra debug information) |
| OPENCV_TEST_REQUIRE_DATA | bool | false | same as `--test_require_data` option (fail on missing non-required test data instead of skip) |
| OPENCV_TEST_CHECK_OPTIONAL_DATA | bool | false | assert when optional data is not found |
| OPENCV_IPP_CHECK | bool | false | default value for `--test_ipp_check` and `--perf_ipp_check` |
| OPENCV_PERF_VALIDATION_DIR | dir path | | location of files read/written by `--perf_read_validation_results`/`--perf_write_validation_results` |
| ⭐ OPENCV_PYTEST_FILTER | string (glob) | | test filter for Python tests |
**Note:** extra FFmpeg options should be pased in form `key;value|key;value|key;value`, for example `hwaccel;cuvid|video_codec;h264_cuvid|vsync;0` or `vcodec;x264|vprofile;high|vlevel;4.0`
| name | type | default | description |
|------|------|---------|-------------|
| ⭐ OPENCV_FFMPEG_CAPTURE_OPTIONS | string (see note) | | extra options for VideoCapture FFmpeg backend |
| ⭐ OPENCV_FFMPEG_WRITER_OPTIONS | string (see note) | | extra options for VideoWriter FFmpeg backend |
| OPENCV_FFMPEG_THREADS | num | | set FFmpeg thread count |
| OPENCV_FFMPEG_LOGLEVEL | num | | set FFmpeg logging level |
| OPENCV_FFMPEG_DLL_DIR | dir path | | directory with FFmpeg plugin (legacy) |
| OPENCV_FFMPEG_IS_THREAD_SAFE | bool | false | enabling this option will turn off thread safety locks in the FFmpeg backend (use only if you are sure FFmpeg is built with threading support, tested on Linux) |
| OPENCV_FFMPEG_READ_ATTEMPTS | num | 4096 | number of failed `av_read_frame` attempts before failing read procedure |
| OPENCV_FFMPEG_DECODE_ATTEMPTS | num | 64 | number of failed `avcodec_receive_frame` attempts before failing decoding procedure |
| OPENCV_VIDEOIO_GSTREAMER_CALL_DEINIT | bool | false | close GStreamer instance on end |
| OPENCV_VIDEOIO_GSTREAMER_START_MAINLOOP | bool | false | start GStreamer loop in separate thread |
| OPENCV_VIDEOIO_MFX_IMPL | num | | set specific MFX implementation (see MFX docs for enumeration) |
| OPENCV_VIDEOIO_MFX_EXTRA_SURFACE_NUM | num | 1 | add extra surfaces to the surface pool |
| OPENCV_VIDEOIO_MFX_POOL_TIMEOUT | num | 1 | timeout for waiting for free surface from the pool (in seconds) |
| OPENCV_VIDEOIO_MFX_BITRATE_DIVISOR | num | 300 | this option allows to tune encoding bitrate (video quality/size) |
| OPENCV_VIDEOIO_MFX_WRITER_TIMEOUT | num | 1 | timeout for encoding operation (in seconds) |
| OPENCV_VIDEOIO_MSMF_ENABLE_HW_TRANSFORMS | bool | true | allow HW-accelerated transformations (DXVA) in MediaFoundation processing graph (may slow down camera probing process) |
| OPENCV_DSHOW_DEBUG | non-null | | enable verbose logging in the DShow backend |
| OPENCV_DSHOW_SAVEGRAPH_FILENAME | file path | | enable processing graph tump in the DShow backend |
| OPENCV_VIDEOIO_V4L_RANGE_NORMALIZED | bool | false | use (0, 1) range for properties (V4L) |
| OPENCV_VIDEOIO_V4L_SELECT_TIMEOUT | num | 10 | timeout for select call (in seconds) (V4L) |
