#-----------------------------------------------------------------------------
# File Name : test_aedat_legacy_timesurface.py
# Author: Emre Neftci
#
# Creation Date : Tue 16 Mar 2021 01:28:22 PM PDT
# Last Modified :
#
# Copyright : (c) UC Regents, Emre Neftci
# Licence : GPLv3
#-----------------------------------------------------------------------------
from torchneuromorphic.utils import plot_frames_imshow, legacy_aedat_to_events
import torchneuromorphic.transforms as transforms
import pylab as plt
import numpy as np
import sys
device = 'cuda'
events = legacy_aedat_to_events(sys.argv[1])
dt = 1000
size = [2, 346, 260]
process_events = transforms.Compose([
transforms.Downsample(factor=[dt, 1, 1, 1]),
transforms.ToCountFrame(T=1000, size=size),
transforms.ToTensor(),
transforms.ExpFilterEvents(tau=100, length=500, device=device)
])
frames = process_events(events)
plot_frames_imshow(np.array([frames.detach().cpu().numpy()]), nim=1)
plt.show()
def create_dataloader(root='data/nmnist/n_mnist.hdf5',
batch_size=72,
chunk_size_train=300,
chunk_size_test=300,
dt=1000,
transform_train=None,
transform_test=None,
target_transform_train=None,
target_transform_test=None,
**dl_kwargs):
ds = [1, 1]
low_crop = [0, 0]
high_crop = [32, 32]
size = [
2,
np.ceil((high_crop[0] - low_crop[0]) / ds[0]).astype('int'),
np.ceil((high_crop[1] - low_crop[1]) / ds[1]).astype('int')
]
print(size)
default_transform = lambda chunk_size: transforms.Compose([
transforms.CropDims(low_crop, high_crop, [2, 3]),
transforms.Downsample(factor=[dt, 1, ds[0], ds[1]]),
transforms.ToCountFrame(T=chunk_size, size=size),
transforms.ToTensor()
])
if transform_train is None:
transform_train = default_transform(chunk_size_train)
if transform_test is None:
transform_test = default_transform(chunk_size_test)
if target_transform_train is None:
target_transform_train = transforms.Compose(
[transforms.Repeat(chunk_size_train),
transforms.toOneHot(10)])
if target_transform_test is None:
target_transform_test = transforms.Compose(
[transforms.Repeat(chunk_size_test),
transforms.toOneHot(10)])
train_d = NMNISTDataset(root,
train=True,
transform=transform_train,
target_transform=target_transform_train,
chunk_size=chunk_size_train)
train_subset_indices = train_d.keys_by_label[:, :100].reshape(-1)
train_dl = torch.utils.data.DataLoader(
train_d,
batch_size=batch_size,
sampler=SubsetRandomSampler(train_subset_indices),
**dl_kwargs)
test_d = NMNISTDataset(
root,
transform=transform_test,
target_transform=target_transform_test,
train=False,
chunk_size=chunk_size_test,
)
test_subset_indices = test_d.keys_by_label[:, :100].reshape(
-1) - test_d.keys[0]
test_dl = torch.utils.data.DataLoader(
test_d,
batch_size=batch_size,
sampler=SubsetRandomSampler(test_subset_indices),
**dl_kwargs)
return train_dl, test_dl
def create_attn_dataloader(root='data/dvsgesture/dvs_gestures_build19.hdf5',
batch_size=72,
chunk_size_train=500,
chunk_size_test=1800,
dt=1000,
transform_train=None,
transform_test=None,
target_transform_train=None,
target_transform_test=None,
**dl_kwargs):
n_events_attention = 100
# default_transform = lambda chunk_size: transforms.Compose([
# transforms.CropDims(low_crop,high_crop,[2,3]),
# transforms.Downsample(factor=[dt,1,ds[0],ds[1]]),
# transforms.ToCountFrame(T = chunk_size, size = size),
# transforms.ToTensor()
# ])
size = [2, 64, 64]
default_transform = lambda chunk_size: transforms.Compose([
transforms.Downsample(factor=[dt, 1, 1, 1]),
transforms.Attention(n_events_attention, size=size),
transforms.Downsample(factor=[1, 1, 4, 4]),
transforms.ToCountFrame(T=chunk_size, size=[2, 16, 16]),
transforms.ToTensor()
])
if transform_train is None:
transform_train = default_transform(chunk_size_train)
if transform_test is None:
transform_test = default_transform(chunk_size_test)
if target_transform_train is None:
target_transform_train = transforms.Compose(
[transforms.Repeat(chunk_size_train),
transforms.toOneHot(11)])
if target_transform_test is None:
target_transform_test = transforms.Compose(
[transforms.Repeat(chunk_size_test),
transforms.toOneHot(11)])
train_d = DVSGestureDataset(root,
train=True,
transform=transform_train,
target_transform=target_transform_train,
chunk_size=chunk_size_train)
train_dl = torch.utils.data.DataLoader(train_d,
batch_size=batch_size,
shuffle=True,
**dl_kwargs)
test_d = DVSGestureDataset(root,
transform=transform_test,
target_transform=target_transform_test,
train=False,
chunk_size=chunk_size_test)
test_dl = torch.utils.data.DataLoader(test_d,
batch_size=batch_size,
**dl_kwargs)
return train_dl, test_dl