Demo code of P&G Hackathon
This repo contains code of the demo presented in P&G Hackathon 2018. Video
Opencv 3.1
Caffe
Nvidia DIGITS
Step 1. Clone and install caffe from this repository
Step 2. Once caffe is installed, go to caffe/examples/cpp_classification/ directory and substitute the classification.cpp file from the one provided in this repository.
Step 3. Clone this repository using the command:
git clone https://github.com/2vin/pg.git
Step 3. Download the model files and extract them in suitable directory.
Step 4. Once everything is done, open the src/main.cpp and edit the corresponding paths in following line:
string str = "../caffe/build/examples/cpp_classification/classification.bin '/home/mouz/Desktop/models/deploy.prototxt' '/home/mouz/Desktop/models/snapshot_iter_100.caffemodel' '/home/mouz/Desktop/models/mean.binaryproto' '/home/mouz/Desktop/models/labels.txt' '/home/mouz/Desktop/PG/code-MV/data/query/query.jpg'";
Step 5. Run ./compile.sh from the root directory of the repository to compile the project.
From the root directory, run the following command:
./bin/main
This will open a screen with "Search". Click on "Search" icon and it will mode to detect mode. Show the photo in required area and click on the "Green Tick" icon. This image will be sent to caffe recognizer and result will be displayed on the screen.
Due to data limitations, the categories are limited to "Cocacola","Vicks","Oral B" & "Duracell" only with around 400 images per category. Once, more data with better quality is added and trained, the results can be improved and scaled easily.
The current accuracy is around 88%, which can be increased with proper dataset. The dataset used for this demonstration can be found here
(Note: This repository contains only the image search part of the whole system. For updates on voice interfacing, check this repository)