Skip to content

Latest commit

 

History

History
32 lines (24 loc) · 1.9 KB

README.md

File metadata and controls

32 lines (24 loc) · 1.9 KB
Raw

Recognition with manual RoI

A web interface for object recognition purposes by providing manual RoI (Region of Interests). Region boundaries are manually specified on the image, instead of utilizing RPN (Region Proposal Network) as in Faster-RCNN. Unlike the Faster-RCNN architecture, a one-dimensional RoI pooling layer is inserted just before the classification block by canceling the whole RPN.

P.S. The input shape of the RoI layer can also be enhanced further.

Dependencies

The requirements.txt file includes the list of all Python libraries that project require, such as django, h5py, keras, numpy, tensorflow or tensorflow-gpu.

Simply install them all using the command below:

pip install -r requirements.txt

Download the imageNet weights for VGG16 vgg16_weights_tf_dim_ordering_tf_kernels.h5, and save it to the directory \static\files_cnn\

Example

Main screen Recognition with a single RoI
Main screen Recognition with a single RoI

Future Works & TODOs

Back-end

  • Create a cnn_model object as a singleton
  • Add a presence check also for the weights file
  • Add a functionality of uploading new image
  • Utilize 'img_src' attribute value instead of default image

Front-end

  • Polish the front-end design
  • Disable the main interface once clicking the button 'Find out'
  • Error thrown in case rectangle-drawing direction is not across the south-east