Diamond Do
Diamond Do
@salcanmor Have you modified the first layer in prototxt as below? name: "MobileNet-SSD" layer { name: "data" type: "Input" top: "data" input_param { shape { dim: 1 dim: 3 dim:...
@salcanmor I made the quantization of Mobilenet-SSD and can run MobilenetSSD on CHaiDNN but the accuracy is too bad. I think depthwise conv was not supported, I am waiting for...
Prototxt and caffemodel I cloned from here https://github.com/chuanqi305/MobileNet-SSD
Can you try this? python XportDNN.pyc --quant_type "Xilinx" --deploy_model ./models/MobilenetSSD_300_deploy.prototxt --weights ./models/MobilenetSSD_300_deploy.caffemodel --quantized_deploy_model ./models/MobilenetSSD_300_quantized_deploy.prototxt --calibration_directory ./data/VOC0712 --calibration_size 32 --bitwidths 8,8,8 --dims 3,300,300 --transpose 2,0,1 --channel_swap 2,1,0 --raw_scale 255.0 --mean_value 127.5,127.5,127.5...
30fps, but the accuracy is too bad. i think it did not run full network
I can process video as below flow Open video file LOOP { grab frame 1 grab frame 2 normaliztion (frame1, frame2, outbuff) xiReadInput(outbuff, input) xiExe(input, output) unpack(output) }
@zhijian-liu did you have training code of ResNet50 on Cifar10 as wrote in the paper?
@zhijian-liu did you have training code of ResNet50 on Cifar10 as wrote in the paper?
@kaustubhharapanahalli I also want to re-train it could you share the train code?
@adambielski could you note the license?