nadavbar/node-cntk-fastrcnn

A node wrapper a CNTK Fast-RCNN model

node-cntk-fastrcnn's Issues

Add support for evaluating models trained with python

The cntk fast rcnn tutorial provides two methods of training models one with brainscript and one with python. The structure of the python model is different than the structure of the brainscript version. This repo currently only supports models trained in brainscript. Can we provide support for models structure like they are in the attached script.

frcnn_detector.py.txt

Add configuration for the classes names and labels

The list of classes and their labels/ids is currently hardcoded in the python script. Add it as a part of configuration file, or setup script.

Add support for DSVM

Currently the module requires cntk and anaconda to be in the same directory. On the DSVM this is not the case is there a way add custom configuration for these paths.

Support latest CNTK version with pyhon 3.5

Add code that searched for the CNTK python env with the latest version and use that version

Add support for a progress indicator event

The implementation will parse the output of the python script for the amount of total files and for the number of processed files and will raise an appropriate event in JavaScript land