Method to improve accuracy of quantized multi-stage object detection network

A method of object detection comprising:

• storing image data of an input image in a memory;
• and
• detecting one or more objects in said input image using a processor to execute a quantized multi-stage object detection network, wherein quantization of said quantized multi-stage object detection network is performed by (i) generating quantized image data by performing a first data range analysis on said image data of said input image, (ii) generating a feature map and proposal bounding boxes by applying a region proposal network (RPN) to said quantized image data, (iii) selecting a plurality of ground truth bounding boxes from said input image, (iv) performing a region of interest pooling operation using said feature map and said plurality of ground truth bounding boxes of said input image instead of the proposal bounding boxes generated by the RPN, (v) generating quantized region of interest pooling results by performing a second data range analysis on results from said region of interest pooling operation performed using said feature map and said plurality of ground truth bounding boxes of said input image, and (vi) applying a region-based convolutional neural network (RCNN) to the quantized region of interest pooling results.