I have modified darknet.prototxt with BN layers and retrained the network with VOC2007 dataset.
I also modified a python script I found for Yolo to support the inference from the newly trained Yolo-v2 model with visualization support of both the ground truth image (if label file exists) and the marked image. The script has non local-maxima suppression for the bounding boxes. I tested on VOC2012 and the result seems to be pretty decent.
The new Caffe model prototxt file and the python script are all included under caffe-yolo-9000/examples/yolo/voc_model.
I have modified darknet.prototxt with BN layers and retrained the network with VOC2007 dataset.
I also modified a python script I found for Yolo to support the inference from the newly trained Yolo-v2 model with visualization support of both the ground truth image (if label file exists) and the marked image. The script has non local-maxima suppression for the bounding boxes. I tested on VOC2012 and the result seems to be pretty decent.
The new Caffe model prototxt file and the python script are all included under caffe-yolo-9000/examples/yolo/voc_model.