def generate_sample(input_model,
weights,
quantized_model,
scaling_mode,
calibration_algo,
conv_algo,
iterations=10,
enable_1st_conv=False,
sampling_single=False):
(blobs, params, top_blobs_map, bottom_blobs_map, conv_top_blob_layer_map,
conv_bottom_blob_layer_map, winograd_bottoms,
winograd_convolutions) = sampling.sample(input_model, weights, conv_algo,
iterations, enable_1st_conv,
sampling_single)
(inputs_max, outputs_max, inputs_min) = sampling.calibrate_activations(
blobs, conv_top_blob_layer_map, conv_bottom_blob_layer_map,
winograd_bottoms, calibration_algo, "SINGLE", conv_algo)
params_max = sampling.calibrate_parameters(params,
winograd_convolutions, "DIRECT",
scaling_mode.upper(), conv_algo)
generate_sample_impl(input_model, quantized_model, inputs_max, outputs_max,
inputs_min, params_max, enable_1st_conv)
return (top_blobs_map, bottom_blobs_map)
def generate_sample(input_model, weights, quantized_model, scaling_mode, calibration_algo, conv_algo, iterations=10, enable_1st_conv=False):
(blobs, params, top_blobs_map, bottom_blobs_map, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, winograd_convolutions) = sampling.sample(input_model, weights, conv_algo, iterations, enable_1st_conv)
(inputs_max, outputs_max, inputs_min) = sampling.calibrate_activations(blobs, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, calibration_algo, "SINGLE", conv_algo)
params_max = sampling.calibrate_parameters(params, winograd_convolutions, "DIRECT", scaling_mode.upper(), conv_algo)
generate_sample_impl(input_model, quantized_model, inputs_max, outputs_max, inputs_min, params_max, enable_1st_conv)
return (top_blobs_map, bottom_blobs_map)
pprint.pprint(cfg)
while not os.path.exists(args.caffemodel) and args.wait:
print('Waiting for {} to exist...'.format(args.caffemodel))
time.sleep(10)
#caffe.set_mode_gpu()
#caffe.set_device(args.gpu_id)
net = caffe.Net(args.prototxt, args.caffemodel, caffe.TEST)
net.name = os.path.splitext(os.path.basename(args.caffemodel))[0]
imdb = get_imdb(args.imdb_name)
imdb.competition_mode(args.comp_mode)
if not cfg.TEST.HAS_RPN:
imdb.set_proposal_method(cfg.TEST.PROPOSAL_METHOD)
if args.quantized_prototxt == None:
test_net(net, imdb, max_per_image=args.max_per_image, vis=args.vis)
else:
(blobs, params, top_blobs_map, bottom_blobs_map, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, winograd_convolutions) = sample_net(args.prototxt, net, imdb, args.sample_iters, args.quant_mode, args.enable_1st_conv_layer)
(inputs_max, outputs_max, inputs_min) = sampling.calibrate_activations(blobs, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, args.calibration_algos, "SINGLE", args.conv_algo)
params_max = sampling.calibrate_parameters(params, winograd_convolutions, "DIRECT", args.quant_mode.upper(), args.conv_algo)
calibrator.generate_sample_impl(args.prototxt, args.quantized_prototxt, inputs_max, outputs_max, inputs_min, params_max, args.enable_1st_conv_layer)
compiled_net_str = caffe.compile_net(args.prototxt, caffe.TEST, "MKLDNN")
raw_net_basename = os.path.basename(args.prototxt)
compile_net_path = "./compiled_" + raw_net_basename
with open(compile_net_path, "w") as f:
f.write(compiled_net_str)
calibrator.transform_convolutions(args.quantized_prototxt, compile_net_path, top_blobs_map, bottom_blobs_map, args.unsigned_range, args.concat_use_fp32, args.unify_concat_scales, args.conv_algo, args.enable_1st_conv_layer)
while not os.path.exists(args.caffemodel) and args.wait:
print('Waiting for {} to exist...'.format(args.caffemodel))
time.sleep(10)
#caffe.set_mode_gpu()
#caffe.set_device(args.gpu_id)
net = caffe.Net(args.prototxt, args.caffemodel, caffe.TEST)
net.name = os.path.splitext(os.path.basename(args.caffemodel))[0]
imdb = get_imdb(args.imdb_name)
imdb.competition_mode(args.comp_mode)
if not cfg.TEST.HAS_RPN:
imdb.set_proposal_method(cfg.TEST.PROPOSAL_METHOD)
if cfg.TEST.PROPOSAL_METHOD == 'rpn':
imdb.config['rpn_file'] = args.rpn_file
if args.quantized_prototxt == None:
test_net(net, imdb, max_per_image=args.max_per_image, vis=args.vis)
else:
(blobs, params, top_blobs_map, bottom_blobs_map, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, winograd_convolutions) = sample_net(args.prototxt, net, imdb, args.sample_iters, args.quant_mode, args.enable_1st_conv_layer)
(inputs_max, outputs_max, inputs_min) = sampling.calibrate_activations(blobs, conv_top_blob_layer_map, conv_bottom_blob_layer_map, winograd_bottoms, args.calibration_algos, "SINGLE", args.conv_algo)
params_max = sampling.calibrate_parameters(params, winograd_convolutions, "DIRECT", args.quant_mode.upper(), args.conv_algo)
calibrator.generate_sample_impl(args.prototxt, args.quantized_prototxt, inputs_max, outputs_max, inputs_min, params_max, args.enable_1st_conv_layer)
compiled_net_str = caffe.compile_net(args.prototxt, caffe.TEST, "MKLDNN")
raw_net_basename = os.path.basename(args.prototxt)
compile_net_path = "./compiled_" + raw_net_basename
with open(compile_net_path, "w") as f:
f.write(compiled_net_str)
calibrator.transform_convolutions(args.quantized_prototxt, compile_net_path, top_blobs_map, bottom_blobs_map, args.unsigned_range, args.concat_use_fp32, args.unify_concat_scales, args.conv_algo, args.enable_1st_conv_layer)
