Live Image Classification

This application runs know CNN image classifiers on live/pre-recorded video stream.

MIVisionX Image Classification Control

MIVisionX Image Classification

Usage

Prerequisites

Build & Install MIVisionX

Build

git clone https://github.com/kiritigowda/MIVisionX-Classifier
cd MIVisionX-Classifier
cmake .
make

Run

Usage: ./classifier 
        --inception <inceptionV4 weights.bin> [optional]
        --resnet50 <resnet50 weights.bin>     [optional]
        --vgg16 <vgg16 weights.bin>           [optional]
        --googlenet <googlenet weights.bin>   [optional]
        --resnet101 <resnet101 weights.bin>   [optional]
        --resnet152 <resnet152 weights.bin>   [optional]
        --vgg19 <vgg19 weights.bin>           [optional]
        --label <label text>                  [required]
        --video <video file>/<--capture 0>    [required]

Generating weights.bin for different Models

Download or train your own caffemodel and run the model_compiler to get the corresponding weights.bin file.

1. Download or train your own caffemodel.

Here is the sample download link that contains all the prototxt:

https://github.com/SnailTyan/caffe-model-zoo

2. Using Model Compiler

To convert a caffemodel into AMD NNIR model:

% python caffe_to_nnir.py <net.caffeModel> <nnirOutputFolder> --input-dims <n,c,h,w> [--verbose <0|1>]

To convert an AMD NNIR model into OpenVX C code:

% python nnir_to_openvx.py <nnirModelFolder> <nnirModelOutputFolder>

The weights file will be generated and you can use that as an input for this project.

--label text

The labels.txt file in this project.

--video file

Test the classification on your own video. Give the path to your video.

--capture 0

If you want to test with a live cam, turn on this option.

Example

./classifier --inception /PATH/TO/inceptionV4/weights.bin --resnet50/PATH/TO/resnet50/weights.bin ...
...
--vgg19 /PATH/TO/vgg19/weights.bin 
--label labels.txt
--capture 0