| import torch |
| import argparse |
| import py3nvml |
| import timeit |
|
|
| parser = argparse.ArgumentParser('Profile the dwt') |
| parser.add_argument('method', choices=['torch', 'numpy'], |
| help='Method to use to calculate dwt') |
| parser.add_argument('xfm', choices=['dwt', 'dtcwt'], |
| help='which transform to use') |
| parser.add_argument('-f', '--forward', action='store_true', |
| help='Only do forward transform (default is fwd and inv)') |
| parser.add_argument('-j', type=int, default=2, |
| help='number of scales of transform to do') |
| parser.add_argument('-s', '--size', default=0, type=int, |
| help='spatial size of input') |
| parser.add_argument('--device', default='cuda', choices=['cuda', 'cpu'], |
| help='which device to test') |
| parser.add_argument('--wave', default='db4', |
| help='which wavelet to use') |
| parser.add_argument('--batch', default=16, type=int, |
| help='Number of images in parallel') |
|
|
| if __name__ == "__main__": |
| args = parser.parse_args() |
| py3nvml.grab_gpus(1) |
| if args.size > 0: |
| size = (args.batch, 1, args.size, args.size) |
| else: |
| size = (args.batch, 1, 128, 128) |
|
|
| if args.method == 'torch': |
| if args.xfm == 'dwt': |
| t = timeit.Timer('ifm(xfm(x))', |
| setup=""" |
| import torch |
| from pytorch_wavelets import DWT, IDWT |
| x = torch.randn(*{sz}).to('{dev}') |
| xfm = DWT(J={J}, wave='{wave}').to('{dev}') |
| ifm = IDWT(wave='{wave}').to('{dev}')""".format(sz=size, dev=args.device, J=args.j, |
| wave=args.wave)) |
| print('5 run average is {:.3f}s'.format(t.timeit(number=5)/5)) |
| else: |
| t = timeit.Timer('ifm(xfm(x))', |
| setup=""" |
| import torch |
| from pytorch_wavelets import DTCWTForward, DTCWTInverse |
| x = torch.randn(*{sz}).to('{dev}') |
| xfm = DTCWTForward(J={J}).to('{dev}') |
| ifm = DTCWTInverse(J={J}).to('{dev}')""".format(sz=size, dev=args.device, J=args.j)) |
| print('5 run average is {:.3f}s'.format(t.timeit(number=5)/5)) |
| else: |
| if args.xfm == 'dwt': |
| t = timeit.Timer('ifm(xfm(x))', |
| setup=""" |
| import numpy as np |
| import pywt |
| x = np.random.randn(*{sz}) |
| xfm = lambda a: pywt.wavedec2(a, '{wave}', level={J}, mode='reflect') |
| ifm = lambda a: pywt.waverec2(a, '{wave}', mode='reflect') |
| """.format(sz=size, wave=args.wave, J=args.j)) |
| print('5 run average is {:.3f}s'.format(t.timeit(number=5)/5)) |
| else: |
| t = timeit.Timer(""" |
| for b in x: |
| for c in b: |
| xfm.inverse(xfm.forward(c, nlevels={J})) |
| """.format(J=args.j), setup=""" |
| import numpy as np |
| import dtcwt |
| x = np.random.randn(*{sz}) |
| xfm = dtcwt.Transform2d(biort='near_sym_a', qshift='qshift_a') |
| """.format(sz=size)) |
| print('5 run average is {:.3f}s'.format(t.timeit(number=5)/5)) |
| |
|
|
