Hi, Some object detection algorithms like MTCNN face detector needs the image to be processed in different scales. Caffe supports the reshaping of input data blob through the function "GraphObject.blobs['data'].reshape(, , , )". Is it possible to do this in Movidius, before feeding the data to the graph?
Hi, thanks for the comment.
It does not tell anything about dynamically changing the input image size (Movidius seems to take in the cropped version of image when changing the image size). This will be a very useful feature, don't you think?
@sahadcp Movidius tools are getting images with various sizes but they are scaled (and cropped if not square) to size which is used when training the network. This means if you use Caffe model and AlexNet for example, scaling will be to 227x227, if GoogLeNet is used size will be 224x224 and nothing different than this.
@georgievm_cms But, some algorithms like RCNN, faster RCNN, MTCNN etc need the image to be fed into the neural net in different sizes. ie, it needs to change dynamically. Not fixed.