fuisz/logo-detection repository overview

⁠Logo Detection

⁠Dependencies

• CUDA 7.5
• Docker
• Python
• Numpy

⁠Datasets

⁠Download and Setup FlickrLogos 27

Source: FlickrLogos 27⁠

cd /data/flickr_27
./download_flickr_logos.sh

⁠Train the Network

run-docker -train -net caffenet -dataset flickr27 

⁠Test the Network

run-docker -test -net caffenet-flickr  

⁠TODO

1. Datasets in the right format to be consumed by Caffe, one zip file for each dataset, along with any scripts used to format the data.

• FlickrLogos 32⁠
• BelgaLogos⁠

1. Fine-tune an existing Caffe network for logo detection using the Fast-RCNN⁠ framework. Ideally experiment with multiple networks.

• Script to train and test each of the networks with each dataset
• Technical document with (README.md) :
  • Step by step instructions on how to setup and replicate the experiments
  • Results of the experiments

1. Create a C++ program or Python script that uses the trained network to detect logos on any given image. The otuput should be: logo_identity,x1,y1,x2,y2