Object detection with video decoding sample

Object detection with video decoding sample#

This sample shows how to run video decoding and object detection using pre-trained YoloV2 Caffe Model

The sample demonstrates the use of mv_compile utility to do video decoding and inference.

The sample has two .cpp files, mvobjdetect.cpp and visualize.cpp. But it needs extra header files, .cpp files, and inference deployment library which is generated by mv_compile utility to completely build and execute the application.

Prerequisites#

Linux
- Ubuntu 20.04/22.04
ROCm supported hardware
- GPU: AMD Radeon™ Graphics [Required]
- APU: AMD Radeon™ Mobile/Embedded [optional]
ROCm
Build & Install MIVisionX
- MIVisionX installs model compiler at /opt/rocm/libexec/mivisionx
- mv_compile installs at /opt/rocm/bin and mvdeploy_api.h installs at /opt/rocm/include/mivisionx
Install MIVisionX Model Compiler Prerequisites
Add MIVisionX libraries & executables to PATH

export PATH=$PATH:/opt/rocm/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/rocm/lib

Sample - Detection Using Pre-Trained CAFFE Model#

Step 1. Download pre-trained YoloV2 caffe model - yoloV2Tiny20.caffemodel #

wget https://github.com/kiritigowda/YoloV2NCS/raw/master/models/caffemodels/yoloV2Tiny20.caffemodel

Step 2. compile model for OpenVX backend using mv_compile utility#

The mv_compile utility generates deployment library, header files, and .cpp files required to run inference for the specified model.

Usage:

mv_compile  --model 	        <model_name: name of the trained model with path> 		[required]
	    --install_folder   <install_folder:  the location for compiled model> 		[required]
	    --input_dims 	<input_dims: n,c,h,w - batch size, channels, height, width> 	[required]
	    --backend 	        <backend: name of the backend for compilation> 	  		[optional - default:OpenVX_Rocm_GPU]
	    --fuse_cba 	<fuse_cba: enable or disable Convolution_bias_activation fuse mode (0/1)> 				[optional - default: 0]
	    --quant_mode       <quant_mode: fp32/fp16 - quantization mode: if enabled the model and weights are converted> 	[optional -default: fp32]

Sample:

mv_compile --model yoloV2Tiny20.caffemodel --install_folder mvdeploy --input_dims 1,3,416,416

NOTE:

There will be a file libmv_deploy.so (under ./lib), weights.bin and mvtestdeploy sample app (under ./bin).
There will be mv_extras folder for extra post-processing helper functions.
Open mvdeploy_api.h to go through API functions supported for inference deployment.

Step 3. Make sure mvtestdeploy utility runs#

mvtestdeploy is a pre-generated application built in Step 3 which shows how to deploy inference for an input image file

Usage:

./bin/mvtestdeploy 	<input data file: image/video>					[required]
			<output file - (.bin)>						[required]
			--install_folder <folder name or '.' for current folder>	[required]
			--t <N: number of iterations>					[required]

Sample:

cd mvdeploy
./bin/mvtestdeploy /opt/rocm/share/mivisionx/samples/mv_objdetect/data/images/img_04.JPG output.bin --install_folder . --t 100

NOTE: This runs inference for an input file and generate output for N number of iterations.

Step 4. Build mvobjdetect example#

mvobjdetect is built on top of all the files generated in Step 4. Basically it shows how to add preprocessing OpenVX nodes for video decoding and image_to_tensor conversion.
Go through mvobjdetect.cpp file.
This exercise uses a single or multiple video streams for input.
The second part of the tutorial shows how to run it through multiple video files.

TODO:

Copy all files in cloned sample folder (mvobjdetect.cpp, visualize.cpp, visualize.h and CMakeLists.txt) into mvdeploy folder. This brings all the files into a single folder to build and run the sample.

cp /opt/rocm/share/mivisionx/samples/mv_objdetect/mvobjdetect.cpp .
cp /opt/rocm/share/mivisionx/samples/mv_objdetect/visualize.cpp .
cp /opt/rocm/share/mivisionx/samples/mv_objdetect/visualize.h .
cp /opt/rocm/share/mivisionx/samples/mv_objdetect/CMakeLists.txt .

Step 5. cmake and make mvobjdetect#

mkdir mv_build
cd mv_build
cmake ../
make -j

Note: if build directory exists from previous build, name the new build directly differently (eg: mv_build).

Step 6. Run object detection with video/image#

Usage:

Usage: mvobjdetect <options>
	<input-data-file: .jpg, .png, .mp4, .m4v>: is filename(s) to initialize input tensor         	[required]
	<output-data-file/- >: for video all frames will be output to single file OR '-'for no output	[required]
	--install_folder <install_folder> : the location for compiled module                         	[required]
	--bb <channels, threshold_c threshold_nms> bounding box detection parameters                 	[required]
	--frames <#num/eof/loop> : num of frames to process inference         	[optional: default till eof, loop: loop the video]
	--backend <backend>: is the name of the backend for compilation  	[optional: default OpenVX_Rocm_GPU]
	--argmax <topK> : give argmax output in vec<label,prob>          	[optional: default no argmax]
	--t <num of interations> to run for performance                  	[optional: default 1]
	--hwdec <1/0>:use hwaccel for decoding                            [optional: default cpu decoding, 1:hwdec 0:cpu dec]
	--label <labels.txt>                                             	[optional: default use yolo_v2 20 classes]
	--v :if specified visualize the result on the input image        	[optional: default no visualization]

Sample

cd ..
./mv_build/mvobjdetect /opt/rocm/share/mivisionx/samples/mv_objdetect/data/images/img_04.JPG - --install_folder . --bb 20 0.2 0.4 --v
./mv_build/mvobjdetect <PATH TO VIDEO FILE> - --install_folder . --bb 20 0.2 0.4 --v

Note: Make sure the input image and video file exist and specified with full path

Step 7. Run object detection with multiple video streams (e.g batch 4, 8 and 16)#

Go through steps 3 to 6, this time compiling the model for a batch of 4.
Also this sample can do batch of 8 and 16 decoding as well.
- For batch 8: --input_dims 8,3,416,416 and create Videos_8.txt file with 8 input streams
- For batch 16: --input_dims 16,3,416,416 and create Videos_16.txt file with 16 input streams

cd ..
mv_compile --model yoloV2Tiny20.caffemodel --install_folder mvdeploy_batch4 --input_dims 4,3,416,416
cd mvdeploy_batch4
cp ../mvobjdetect.cpp ../visualize.cpp ../visualize.h ../CMakeLists.txt .
mkdir mv_build4
cd mv_build4
cmake ../
make -j
cd ..
./mv_build4/mvobjdetect <Videos_4.txt> - --install_folder . --bb 20 0.2 0.4 --v

Note:

Where Videos_4.txt has the names of input video files with full path.
e.g: Use the ../data/Videos/Videos_4.txt file. Modify it to specify your input files.
The last bit (:0 / :1) chooses software or hardware mode for decoding.

Step 10. Sample output for multiple video object detection#

License#

This project is licensed under the MIT License - see the LICENSE.md file for details

Author#

Rajy Rawther - mivisionx.support@amd.com